COO-3077-151
Courant Mathematics and
Computing Laboratory
U.S. Department of Energy
The PL/I Programming Language
Paul Abrahams
Research .\uc\ Development Re|X)rt
Prepared under Contract EY-76-C-02-3077
with the Office of Energy Research
Mathematic s cVk\ Computing
March 1978
New York University
UNCLASSIFIED
Courant Mathematics and Computing Laboratory
New York University
Mathematics and Computing COO-3077-151
THE PL/I PROGRAMMING LANGUAGE
Paul Abrahams
March 1978
U. S. Department of Energy
Contract EY-76-C-02-3077
UNCLASSIFIED
This report is to appear as an article on the PL/I
programming language in the Encyclopedia of Computer
Science and Technology published by Marcel Dekker, Inc,
11
TABLE OF CONTENTS
Page
INTRODUCTION 1
Syntactic Conventions 6
DATA TYPES 9
Arithmetic Types 9
String Types 11
Pictured Types 13
Pointers , Areas and Offsets 19
Files 20
Labels 22
Entries 23
Formats 24
Arrays 25
Structures 26
DECLARATIONS 29
Manifest, Explicit, Contextual
and Implicit Declarations 29
Declarations of Statement-Names 31
Attribute Consistency and Completeness 32
Standard and User Defined Defaults 38
The LIKE-Attribute 41
EXPRESSIONS, TYPE CONVERSION, AND ASSIGNMENT 43
Prefix and Infix Expressions 45
Builtin Functions 51
Type Conversion 64
Promotion 68
The Assignment-Statement 69
STORAGE TYPES 7 2
Static Storage 72
Automatic Storage 73
Controlled Storage 73
Based Storage 74
The Refer-Option 77
Lef t-to-Right Correspondence 79
Allocation in Areas 80
Parameter Storage 81
Defined Storage 82
Alignment 84
Initialization 86
111
PROCEDURES, SCOPES AND ENVIRONMENTS 8 8
The RETURN-Statement 90
Arguments and Parameters 91
Options 9 3
Recursion 94
The GENERIC-Attribute 9 5
Blocks and Scopes 96
Internal and External Scope 9 9
Entry Values and Environments 100
ON-UNITS AND ON-STATEMENTS 102
The ON-Statement, REVERT-Statement ,
and SIGNAL-Statement 10 5
Enablement and Disablement 10 8
Builtin Functions for ON-Conditions 109
Categorization of the ON-Conditions 110
OTHER STATEMENTS AFFECTING FLOW OF CONTROL 115
Conditional Statements 115
The DO-Statement 116
The GOTO-Statement 119
The STOP-Statement and the Null-Statement 121
FILES AND RECORD INPUT-OUTPUT 122
File Attributes 122
File Opening and Attribute Determination 123
File Closing 127
Operations on Record Files 127
STREAM INPUT-OUTPUT 135
Data Lists 137
List-Directed Input-Output 138
Data-Directed Input-Output 140
Edit-Directed Input-Output 142
BIBLIOGRAPHY 150
IV
INTRODUCTION
PL/I is a large and powerful multipurpose prograiraning
language. The intent of the designers of PL/I was to create
a language that could be used in business and in scientific
applications, as well as in systems programming applications
such as writing operating systems. The original design was
developed in 196 3 by a committee of people drawn from IBM
and from SHARE, an IBM user group. For a long time the only
important implementations of PL/I were those developed by IBM
on the 360 and 370 computers, and the implementation on the
GE 645 at the MULTICS Project at MIT. However, during the
early 1970' s a number of other implementations arose. The
implementation of PL/I by other organizations was given impetus
by the development of a national and international standard
for PL/I by a subcommittee of the American National Standards
Institute, in conjunction with a similar subcommittee of the
European Computer Manufacturers' Association.
The definition of Standard PL/I was formally released
late in 1976, but the content of the standard was publicly known
well before then. The standard itself was written in a novel
manner as a set of algorithms, expressed in highly stylized
English, for the operation of a hypothetical PL/I machine.
The version of PL/I described in this article is Standard PL/I.
The design of PL/I drew heavily on the major languages
that existed in 1963: Fortran, Cobol , and Algol 60. The
syntax of PL/I most resembles that of Fortran, but without
Fortran's rigid rules for program formatting. The notion of
block structure was taken from Algol 60, while PL/I structures
were taken from the record descriptions of Cobol. However, a
great many features were added to PL/I that have no counter-
part in its ancestor languages.
An example of a PL/I program is given in figure 1. A
program is written as a sequence of external procedures , which
are defined in such a way that they can be compiled separately
and then linked together when the program is executed. Within
an external procedure, there can be internal procedures .
In this example, there is one internal procedure, named
GET_DIGRAM. Each procedure in the program constitutes a block,
In addition, a block can be delimited by the PL/I statements
BEGIN and END (as in Algol 60) .
The internal procedure GET_DIGRAM communicates with the
outer procedure DIGRAMS via arguments passed to GET_DIGRAM
by the CALL statement. From the viewpoint of GET_DIGRAM,
these arguments appear as parameters and are listed in the
PROCEDURE- statement .
The variables used in this program are given in the
DECLARE-statements. In general, a variable (or other use of
an identifier) is described by a set of attributes . Not all
of these need be given in the DECLARE- statement; those that
are not given are deduced through the application of a set of
defaulting rules. In fact, defaulting is applied in a great
many contexts within PL/I.
On account of its comprehensive nature, PL/I is a diffi-
cult language to learn in its entirety. For that reason it
was designed so that a user could learn just those parts of
the language that he needed, and ignore the rest of it until
the occasion arose to use some previously untried feature.
The extensive defaulting conventions were included, for a
large part, to make it possible to write programs without
having to learn about obscure and irrelevant attributes. For
instance, one can write business programs in PL/I without
ever realizing that the language includes complex numbers and
an extensive repertoire of mathematical builtin functions.
Since Standard PL/I is intended to be implemented on a
variety of machines, the standard provides that a number of
characteristics of the language are implementation-defined.
For example, machines differ in their word lengths; therefore
the maximum number of digits that need be carried in a
floating point conputation is left implemented-def ined . In the
description of PL/I given in this article, implementation-defined
features of the language are referred to frequently.
As of this writing, a standard subset version of PL/I
is under development by the American National Standards Insti-
tute. The definition of this standard subset will probably be
released by the time that this article appears. Moreover, an
extension of the subset to include facilities for real time
and concurrent programming is also under development by the
same group.
Figure 1. A Sample PL/I Program
/* THIS PROGRAM READS IN A TEXT AND COUNTS THE NUMBER OF TIMES
THAT EACH ALPHABETIC DIGRAM OCCURS. A DIGRAM IS A SEQUENCE
OF TWO ADJACENT CHARACTERS. FOR INSTANCE, THE DIGRAMS IN
'GRUNGE" ARE GR, RU , UN, NG , AND GE .
*/
DIGRAMS: PROCEDURE OPTIONS (MAIN) ;
DECLARE COUNT(26,26) FIXED(4);
/* COUNT (I, J) GIVES THE CURRENT COUNT OF OCCURRENCES
OF THE DIGRAM FORMED FROM THE I-TH LETTER AND
' THE J-TH LETTER.
/*
DECLARE ALPH CHARACTER (26 ) INITIAL
( 'ABCDEFGHIJKLMNOPQRSTUVWXYZ' );
DECLARE (P1,P2) FIXED; /* ALPHABETIC POSITION OF LETTER */
DECLARE (L1,L2) CHARACTER ( 1 ) ;
/* FIRST AND SECOND LETTERS OF DIGRAM */
DECLARE DONE BIT(l) INITIAL (' 0 ' B) ;
/* COMPLETION FLAG */
/* SET ALL ELEMENTS OF THE COUNT ARRAY TO ZERO */
COUNT = 0;
/* READ AND PROCESS DIGRAMS */
RDLOOP: DO WHILE ('I'B); /* DO FOREVER */
CALL GET_DIGRAM(L1,L2,D0NE) ;
IF DONE THEN
GO TO PRINT;
PI = INDEX (ALPH, LI ) ;
P2 = INDEX (ALPH, L2) ;
/* N.B.— INDEX RETURNS ZERO IF LETTER NOT IN ALPHABET */
IF PI * P2 > 0 THEN /* DIGRAM IS ALPHABETIC */
C0UNT(P1,P2) =C0UNT(P1,P2)+1;
END RDLOOP;
/* PRINT THE RESULTS */
PRINT: DO PI = 1 TO 26;
DO P2 = 1 TO 26;
IF COUNT (PI, P2)>0 THEN /* DIGRAM APPEARED */
PUT EDIT(SUBSTR(ALPH,P1,1) ,
SUBSTR(ALPH,P2,1) , COUNT (PI ,P2) )
(SKIP, 2 A(l) , X(2) , F(4)) ;
END;
END ;
STOP; /* END EXECUTION OF PROGRAM */
Figure 1. Continued
/* INTERNAL PROCEDURE TO EXTRACT THE NEXT PAIR
FROM THE INPUT TEXT
/*
GET_DIGRAM: PROCEDURE (LI , L2 , FLAG) ;
DECLARE (LI, L2) CHARACTER ( 1 ) ; /* LETTER PAIR */
DECLARE FLAG BIT(l); /* END-OF-DATA INDICATOR */
DECLARE CARD CHARACTER ( 80 ) ; /* INPUT LINE IMAGE */
DECLARE POSN FIXED STATIC INITIAL(80);
/* CHARACTER POSITION IN INPUT
CARD */
DECLARE SYS IN RECORD INPUT FILE;
/* INPUT READ FROM FILE SYSIN */
ON ENDFILE (SYSIN) /* WHEN INPUT EXHAUSTED */
GO TO INPUT_FINISHED;
IF POSN > 79 THEN DO ;
READ FILE(SYSIN) INTO(CARD); /* READ A CARD */
POSN =1; /* PROCESS FROM START OF CARD */
END;
LI = SUBSTR(CARD, POSN, 1);
L2 = SUBSTR(CARD, POSN+1,1);
POSN = POSN+1; /* MOVE TO NEXT PAIR */
RETURN; /* EXIT FROM THIS PROCEDURE */
/* COME HERE IF THE READ STATEMENT ENCOUNTERED AN END OF FILE */
INPUT_FINISHED:
FLAG = 'I'B; /* SIGNAL COMPLETION TO CALLER */
RETURN ;
END GET_DIGRAM;
END DIGRAMS;
Syntactic Conventions
A PL/I external procedure consists of a sequence of
statements . With the exceptions of the IF- statement and the
ON- statement , every statement is followed by a semicolon.
The program is presented in free field format, i.e., statements
do not occupy a fixed position on the line. In fact, line
boundaries are ignored altogether, so a statement can be split
over several lines, or several statements can occupy a single
line. With the exception of the assignment statement and the
null statement, the type of a statement is indicated by the
keyword with which it begins.
A statement, in turn, is written as a sequence of tokens
each of which may be either a delimiter or a nondelimiter.
The types of tokens are :
Delimiters
operator
period
comma
left or right parenthesis
colon
semicolon
text inclusion
Nondelimiters
identifier
arithmetic constant
string constant
isub [discussed below]
Two adjacent nondelim.iters must have at least one blank
between them, and other than that adjacent tokens may have
any number of blanks between them. For example the statement
DO IVAL = Q TO (A + 3) ;
could be written more compactly as
DO IVAL^Q T0(A+3) ;
but it could not be further condensed to
D0IVAL=QT0(A+3) ;
The last example is in fact a valid PL/I statement with an
entirely different meaning. A comment may be used in any
place where a blank can appear. A comment is written as the
characters ■■/*" followed by a sequence of characters
not containing " */" , followed by " */" , e.g.,
/* THIS IS A COMMENT */
An identifier consists of a letter followed by any number of
letters, digits, and break characters "_" , e.g.,
P0PE_LE0_THE_15TH. The syntax of most of the other kinds of
tokens is discussed below.
A text inclusion has the form
%INCLUDE textname ;
The textname refers to an externally stored piece of text,
which replaces the text inclusion when the program is trans-
lated. On account of the variation in operating environments
on different machines, the interpretation of the textname is
implementation defined. With this facility, it is possible
to use the same version of a chunk of program in many differ-
ent external procedures, even ones written by different
programmers .
Keywords can be used as identifiers; in this respect PL/I
differs from COBOL, which treats keywrods as reserved words,
7
and from ALGOL, which uses a distinct typeface to represent
them. For example,
IF IF THEN THEN = ELSE ;
is a valid sequence of statements in PL/I. The first statement
is an IF statement that tests the variable IF; the second is
an assignment statement whose target is the variable THEN.
Most of the long key words can be abbreviated, e.g. , CTL for
CONTROLLED or NOFOFL for NOFIXEDOVERFLOW. The names of some
builtin functions can also be abbreviated.
There are three different kinds of PL/I statements that
head groups of statements: the DO statement, the PROCEDURE
statement, and the BEGIN statement- For all three, the group
is ended by an END statement. If a statement name is attached
to the statement that heads a group, then the same statement
name can also be attached to the END statement that terminates
the group, thus indicating which statement is closed out by
the END. A single END statement can close out more than one
group, however, as the following example illustrates:
ALEPH: PROCEDURE;
DO;
BEIT: BEGIN;
END BEIT;
END ALEPH;
The final END statement closes out both the leading PROCEDURE
statement and the DO statement that follows it.
DATA TYPES
The different kinds of data in PL/I can be classified
into groups called data types, or simply types. The avail-
able types are either aggregate types or saalar types. An
aggregate is composed from simpler types, and can be either
an array or a structure. Arrays and structures are discussed
below. The scalar types can be grouped into printable and
nonprintable types, sometimes known as computational and
nonaomputational. For each type, there can be variables and
values of that type; for some types there can also be
constants .
A constant associates a naime with a single unchanging
value, while a variable associates a name with a location
where a value can be stored. The value of a variable is in
general time-dependent. Variables are introduced into the
program by DECLARE-statements , e.g.,
DECLARE LETTER_SEQUENCE CHARACTER (15) VARYING;
which declares the variable LETTER_SEQUENCE to have character
strings of length from 0 to 15 as its values. However, a
variable can be declared even though no DECLARE- statement is
written for it (see "Declarations" below) , and certain kinds
of constants are also introduced through DECLARE-statements.
Arithmetic Types
The printable types consist of the arithmetic types and
the string types. The arithmetic types are characterized by
four kinds of attributes: the base (binary or decimal) ,
the scale (fixed or float) , the mode (real or complex) and
the preoision. Since all combinations of base, scale, and
mode are permitted, there are eight arithmetic types, pre-
cision aside. The precision of a fixed type consists of a
number- of- digits and a scale- factor ; that of a float type
consists just of the number- of- digits. For example, the
type REAL FIXED DECIMAL (6, 2) (the "(6,2)" indicates the
precision) contains values of the form + DDDD.DD, where the
D's are decimal digits. If the scale factor is omitted, it
is taken as zero. The binary types are similar, except
that binary rather than decimal digits are used. The fixed
and float types correspond to the fixed point and floating
point arithmetic data available on most computers. FIXED is
a generalization of the integer type found in a number of
other programming languages, since the integer type does not
provide for scaling. FLOAT corresponds to the real type of
other languages. A notable feature of the float types is
that they can be used to express the desired accuracy of a
numerical computation independently of the word length of
the computer carrying out the computation.
The type of an arithmetic constant is indicated by its
form. A real fixed decimal constant consists of a sequence
of decimal digits with an optional decimal point and sign,
e.g., - 17.76 . Fixed constants can be scaled; for instance,
-4
78F-4 has the value 0.0078, obtained by multiplying 78 by 10
A real float decimal constant consists of a real fixed decimal
constant followed by an exponent part indicated by the letter E,
10
e.g., 4.832E+12. Binary constants are formed similarly,
except that only binary digits are used, and the number is
followed by the letter B, e.g., 10. IB or 11E16B (designating
3 X 10 ) . Complex constants do not exist as such; a complex
constant is formed as the sum of a real constant and an imag-
inary constant, e.g., 4+31.
String Types
The string types are aharaoter and bit, each of which
in turn may be varying or nonvarying . However, string values
are sequences of characters or bits, and the varying and non-
varying attributes are not applied to them. Each string type
has a maximum length associated with it; for the nonvarying
types, the actual length is always equal to the maximum
length. For instance, the type CHARACTER (14 ) VARYING describes
character strings whose length varies from 0 to 14 characters,
while the type BIT (8) NONVARYING (NONVARYING is the default)
describes bit strings that are always exactly 8 bits long.
A character-string constant is written as a sequence of
characters enclosed in single quotes, with internal quotes
doubled, e.g.,
'THE FARMER' 'S DAUGHTER'
The null character string, which contains no characters, is
written as ' ' . A bit-string constant is written as a sequence
of binary digits enclosed in single quotes and followed by B,
e.g., 'lOlOOl'B; the null bit string is written as ''B. Bit
strings can also be written in base-4, base-8, or base-16
11
notation. For instance, '7400'B3 indicates the base-8 (octal)
constant 7400 (equivalent to ' 111100000000 ' B) , while 'A81'B4
indicates a base-16 constant. (The digit after the B indicates
a power of 2.) This extended notation is not available for
binary arithmetic constants. The one-bit values 'I'B and 'O'B
are particularly useful, as they are the results returned by
the PL/I comparison operators; 'I'B represents true and 'O'B
represents false.
When a string, either bit or character, is declared, the
maximum length need not be given by a constant, so that the
declaration
DECLARE NEWSTR CHARACTER (Kl+2 ) ;
is permissible. The expression Kl+2 must be well-defined at
the time that NEWSTR is created. Strings that appear as
parameters of procedures may have their maximum lengths given
by * , e . g . ,
DECLARE PARAM_3 CHARACTER (*) ;
In this case, the maximum length of PARAM_3 is determined by
the argument corresponding to PARAM_3 . Strings used as para-
meters must have their maximum lengths given either by * or
by an expression composed purely of constants; more general
expressions are not permitted (but are not particularly useful
in this context in any case) .
12
Pictured Types
The pictured types are derived from similar types in COBOL,
but are more general, A pictured type has an associated picture,
e.g., 999V. 99, that describes the appearance of the values of
that type. The values are represented as character strings, and
the semantics of PL/I are such that an implementation is actually
obliged to store them that way. There are no constants of
pictured type. Pictures can be used in input-output formats as
well as in declarations; an example of the declaration of a
pictured type is
DECLARE SALARY PICTURE ' $$$ , $$$V . $$ ' ;
A pictured type has a picture associated with it. The
picture is given by a character string. Within the picture,
parenthesized counts can be used to indicate repeated characters,
so that the picture •$$$$$' can also be written as '(5)$'.
A picture can be either a character picture or a numeric
picture. A numeric pictured type contains, in addition to the
picture itself, a mode specification (either REAL or COMPLEX) .
Character pictures are rather simple. They consist of
just the characters A, 9, and X. The character A stands for
a letter or a blank; the character 9 stands for a digit or a
blank; and the character X stands for anything. Character
pictures are used to validate strings, i.e., to insure that
they are in the proper form. Thus, if we have the declaration
DECLARE STRING_TEST PICTURE 'OJAXXg';
we can assign to STRING_TEST values consisting of three letters
13
(or blanks) followed by any two characters followed by a
single digit (or blank) . If the assigned value does not
have these characteristics, an error will be signalled.
A niimeric picture is one that contains a character other
than A, 9, or X. By this definition, a picture consisting of
all 9's is a character picture rather than a numeric picture.
A numeric pictured type has an associated arithmetic type,
which is determined by the form of the picture together with
the mode. The picture itself is independent of the mode.
Thus if we have the declarations
DECLARE RPIC REAL PICTURE 'ZZZ';
DECLARE CPIC COMPLEX PICTURE 'ZZZ';
RPIC has an associated arithmetic type of REAL FIXED DECIMAL (3 , 0) ,
while CPIC has an associated arithmetic type of COMPLEX
FIXED DECIMAL(3, 0) . (The associated arithmetic type is neces-
sarily decimal.) If the picture contains either of the
characters E or K, it is a float picture (i.e., its associated
arithmetic type is float) ; otherwise, assuming it is numeric,
it is a fixed picture. The associated arithmetic type can be
thought of as specifying the meaning of the values, as distinct
from the representation of the values. The meaning becomes
important when pictured values are used in arithmetic operations.
When a numeric value is assigned to a pictured variable,
the value is edited to conform to the picture. The characters
in the picture determine how the editing is to be done, assuming
that the value has already been converted to the associated
14
arithmetic type. Editing by means of a picture is illustrated
by the following example: Suppose that the value 34.8 is to
be edited using the picture S9999V.99. Then the pictured value
will be +0034.80. In this example the S indicates an explicit
sign, the 9's indicate explicit digits, the V indicates an
implicit decimal point (used to align the numeric value with
the characters of the picture) and the period is an insertion
character. The meanings of the different picture characters
are given in Table 1, and their use is illustrated in Table 2.
The I, R, and T characters represent digits with sign-
overpunching, i.e., the sign of the entire value is combined
with the digit to form a single character. Sign-overpunching
is the standard input convention for COBOL; on a keypunch,
the characters are formed by punching both a sign (11-row for -,
12-row for +) and a digit in a single column. Although CR and
DB are two characters rather than one, the two characters
always go together, and are used to indicate a negative
quantity. CR stands for "credit" and DB for "debit".
The drifting characters $, Z, +, -, and S are used to edit
leading zeros into blanks. When a sequence of drifting charac-
ters appears (they must all be the same one) , the character
drifts to the position to the left of the leading nonzero digit
in the value, and the remaining positions to the left are
blanked out. Any insertion characters within the blanked-out
positions are themselves blanked out.
The period is a true insertion character, in that its
presence or absence has no effect on the value represented by
15
the picture. The period need not appear next to the V, even
though the sequence "V." is often used. A single $, S, +, or -
can be placed at the beginning or at the end of a picture, in
which case it signifies an explicit insertion rather than a
drifting position. For instance, the value 92 edited by the
picture •(5)$V.$$S' yields the string '\6}6$92.00+' ; in this
case the S causes a sign to be inserted, and is not a drifting
character.
Although pictures are ordinarily used to edit real values,
they can be used to edit complex values also. When a variable
is declared to be pictured and complex, the picture is used
to edit both the real and imaginary parts of any complex value
assigned to the variable, and the two parts are concatenated.
However, there is no way to use pictures to insert "I" into
the edited representation of a complex number.
16
Table 1. Meaning of Picture Symbols
(a) Character-picture symbols
A alphabetic character
9 digit or blank permitted
X anything permitted
(b) Numeric-picture symbols
9 digit
Y digit with zero mapped to blank
Z digit with zero-suppression
$ drifting or inserted dollar sign
* drifting asterisk (check protection)
+ drifting or inserted sign for positive values
drifting or inserted sign for negative values
S drifting or inserted sign for all values
CR credit symbol, inserted for negative values
DB debit symbol, inserted for negative values
I digit with positive value indicated by overpunch
R digit with negative value indicated by overpunch
T digit with sign always indicated by overpunch
inserted period
, inserted comma
B inserted blank
/ inserted slash
V implied decimal point
E start of exponent, E inserted
K start of exponent, nothing inserted
17
Table 2. Examples of Numeric Pictures
Picture
Numeric Value
Pictured String
999
7
007
9 9V9
7
070
YY/YY/YY
760404
76/)zJ4/)z54
ZZZZ
23
)z5)zJ2 3
Z,ZZZ
123
]6]6123
Z,ZZZ
1234
\61,234
Z.VZZS
-1.6
1.60-
Z.VZZ+
.03
)zJlz503+
ZV.ZZ+
.03
]6.03 +
-ZZZZS
0
\6fo]6]6ii6jp
-ZZZZ$
2
]6]6m2$
+ZZZZ$
2
+)6]6]62$
ZZ99
0
]6\600
$$$v.$$
1.23
b$1.23
$$B$$$
2345
$2}zJ345
$$B$$$
345
]6]6$345
+++
6
]6+6
6
ms
sss
6
]6+e
$***Y^ **
.07
$***.07
$***^Y**
.07
5****07
$$$99CR
8
]6]6$08W
$$$99CR
-123
Jz$$123CR
$$$99DB
-123
tJ$123DB
999T
71
07A
999T
-71
07J
1999
1776
A776
1999
-1776
1776
ZZZR
71
71
ZZZR
-71
7J
99.V999BKS99
123456
12.346)z$+04
V.99999E99
123
.12300E03
999V9F3
12345
0123
99999F-2
12.34
01234
Note: A indicates
1 with postiive
overpunch.
J indicates 1 with negative overpunch,
18
Pointers, Areas and Offsets
The nonprintable types of data in PL/I are pointers, areas,
offsets, files, labels, entries, and formats. A pointer can
be thought of as the location of a piece of data; it resembles
the ref (reference) notion of Algol 68. However, pointer
variables in PL/I are untyped; that is, a pointer variable
can contain a pointer to data of any type whatsoever. The
only pointer constant is the null pointer, which does not
point at anything and therefore does not corpare equal to any
pointer to an existing object. The null pointer is obtained
as the value of the builtin function NULL of no arguments.
Pointers are used in conjunction with based variables, which
act as templates for an area of storage. Based variables are
discussed below.
An area is a region in which space for based variables
can be allocated. Areas can be cleared of their allocations
in a single operation, thus allowing for wholesale freeing.
Moreover, areas can be moved from one place to another by
means of assignment to area variables, or through input-output
operations. There is one area constant, the empty area,
which is obtained as the value of the builtin function EMPTY
of no arguments. Assignment of the empty area to any area
variable clears the area of its allocations. More precisely,
the old value of the variable is destroyed (though a copy may
exist elsewhere) , and the new value is an area with nothing
allocated in it. The declaration of an area specifies (at
19
least by default) an area size in implementation-defined units
(bytes, words, etc.), e.g.,
DECLARE STRUCTURE_AREA AREA ( 2000 );
When an area is moved, pointers to objects within the area
lose their validity. Therefore, PL/I also provides offsets,
which are pointers relativized to the origin of a given area.
When an area is moved, the offsets of the objects within the
area remain unchanged. Conceptually, pointers and offsets are
related by the equation
pointer = offset + area
but in an actual implementation that equation need not hold.
There is one offset constant, the null offset, which is obtained
by converting the null pointer to an offset.
To make it easier to work with offsets, it is possible
to declare an offset with an implicit area association (which
can be overridden), e.g.,
DECLARE 0FF_FR0M_A3 OFFSET (A3);
DECLARE A3 AREA(300);
When 0FF_FR0M_A3 is referenced in a context where a pointer is
required, the offset value in 0FF_FR0M_A3 is converted to a
pointer relative to the area A3.
Files
A file is, conceptually, a port through which communication
is established between the program and a dataset . A dataset,
in turn, is a collection of information residing on an external
20
medium, accessible to the program only through input-output
operations. During the course of execution of a program,
a given file may be connected to different datasets , or to
no dataset, at different times. Input-output operations
reference a file, which must be connected to an appropriate
dataset; the operations then take place on the dataset. A file
connected to a dataset is said to be open; one not connected
to a dataset is said to be alosed.
The file itself is a file value; each file value is
uniquely associated with a file constant, declared, for example,
by
DECLARE FILECON FILE CONSTANT;
File variables are declared similarly, e.g., by
DECLARE FILEVAR FILE VARIABLE;
If neither CONSTANT nor VARIABLE is specified in the declaration,
the usual default is CONSTANT. Moreover, declarations of file
constants are introduced implicitly in a number of contexts,
so that in practice the programmer rarely needs to write these
declarations. For instance, the PL/I statement
PUT LIST(A,B) ;
causes the values of A and B to be written onto the dataset
associated with the file named SYSPRINT (assumed since no other
file was specified in the PUT-statement) . If no declaration is
explicitly given for SYSPRINT, the declaration
DECLARE SYSPRINT FILE CONSTANT:
21
is assumed. When the PUT-statement is executed, the file
SYSPRINT is opened as a PRINT file (assuming it is not already
open) .
Labels
A label is a name attached to an executable statement so
that control can be transferred to that statement by means of
a GOTO- statement . A label value has two components: a designator
(such as an address) of the statement named by the label, and
an environment , which records the state of execution of the
program at the time when the block containing the label was
entered. The environment is necessary because the address by
itself does not always provide sufficient information to deter-
mine unambiguously the state of execution after a GOTO- statement
has been carried out (see the section "Entry Values and Environ-
ments" below) .
Label constants are declared by the appearance of a label
as a statement-name , as in
BOOK_FOUND: VOLUME = FOLIO (J) ;
which declares BOOK_FOUND as a label constant. In fact, label
constants cannot be declared in any way other than by their
appearance as statement-names. Not all statement-names declare
label constants, however; some of them declare entry constants
and format constants. The type of constant declared by a
statement-name is determined by the type of statement to which
it is attached. Label variables are declared using the
attribute LABEL, e.g.,
22
DECLARE LABVAR LABEL VARIABLE;
A statement-name can be written with one or more subscripts,
and by this convention constant arrays of labels can be created.
For instance, if a block contains executable statements with the
statement-names CASE(-l), CASE(O), CASE(l), and CASE (2), the
appearance of these statement-names constitutes a declaration
of CASE as a constant array of labels, whose single subscript
has a lower bound of -1 and an upper bound of 2 . It is quite
permissible to attach several of these subscripted statement-
names to a single statement, to give them in nonincreasing
order, to attach them to statements also bearing nonsubscripted
statement-names, or even to omit some index values from the set.
For instance, if CASE(O) were omitted from the above set, the
set would still be valid, but a transfer to CASE(O) would be
in error. Typically, an element of a constant array of labels
is selected by a statement such as
GO TO CASE(CASE_NUMBER) ;
Entries
An entry is an entry point to a procedure (see "Procedures,
Scopes, and Environments" below) treated as a datum. An entry
constant is declared by the appearance of a statement-name
attached to a PROCEDURE-statement or an ENTRY-statement . For
instance,
PROCESS_NAME: PROCEDURE (NAME, SPECS) ;
declares PROCESS_NAME as an entry constant, whose associated
23
value is the (single) entry point to the procedure headed by
the PROCEDURE-statement. Entry values, like label values,
carry environments with them. Entry variables and arrays of
entry constants are available. A typical application of an
entry variable arises in writing a procedure to integrate an
arbitrary function; within the procedure, the function is
declared as an entry variable (and a parameter) , and the
actual function to be integrated is passed as an argument.
When an external procedure references an entry in
another external procedure, then the first procedure must
declare the second explicitly as an entry constant, as in
DECLARE SEARCHVAL EXTERNAL ENTRY (CHARACTER (*) )
RETURNS (FIXED) ;
The CONSTANT attribute is assuined by default in this case.
The procedure containing this declaration expects SEARCHVAL
to be an entry to an external procedure, whose expected argu-
ment is a character string of unspecified length, and which
returns a fixed value.
Formats
A format is used to specify the form of data on a dataset
accessed through a stream file (see "Edit-Directed Input-Output"
below) . A format constant is declared by the appearance of a
statement-name on a FORMAT- statement , e.g.,
FMT3: FORMAT (SKIP, 3A, X (M) , A) ;
As with labels and entries, format variables and arrays of
format constants are included in PL/I. Format variables
24
are declared using the attribute FORMAT, e.g.,
DECLARE FMTVAR FORMAT VARIABLE;
Since formats can contain references to variables, e.g., the
M in the example FMT3 above, format values carry environments.
Arrays
Two types of aggregates are provided in PL/I: arrays
and struGtures . An array is a collection of elements all having
the same type; a particular member of the collection is selected
using an appropriate sequence of subscripts. A structure, on
the other hand, is a collection of elements having possibly
different types; a particular member of the collection is
selected by using an appropriate name as the selector. A
powerful feature of PL/I is that it allows aggregates to be
treated as data objects in most contexts, so that it is
possible, for instance/ to add two arrays in a single opera-
tion, or to write a procedure that returns a structure as its
value .
An array is characterized by a sequence of dimensions;
the dimensionality of the array is the number of its dimensions.
Each dimension has a lower bound and an upper bound, and these
can be arbitrary integers. For example, the declaration
DECLARE MESH(-100: 100,200) FLOAT BINARY(40);
declares MESH to be a two-dimensional array. The first sub-
script has lower bound -100 and upper bound 100, while the
second has lower bound 1 (assumed since none is given) and
25
upper bound 200. When reference is made to an element of an
array, the subscripts can be arbitrary expressions, as long as
the values of those expressions can be converted to integers.
As with character strings, bounds can be given by expres-
sions as well as constants. For an array parameter, a pair of
bounds (not a single one) can be given by *. The * can also
be used in a quite different sense to indicate a cross-section
of an array. For instance, MESH (3 ,* ) designates a one-dimais ional
array, with lower bound 1 and upper bound 200, consisting of
those elements of MESH whose first subscript is 3. Any number
of the siabscripts in an array reference can be replaced by * ' s ;
the dimensionality of the resulting array is the number of *'s.
Array variables take on array values . Array values can
arise during the evaluation of an expression, e.g., when two
arrays are added together. Aside from labels, entries, and
formats, there are no array constants in PL/I.
Structures
A structure is a collection of named elements, each of
which can itself be a structure. An example of a two-level
structure declaration is
DECLARE
1 EMPLOYEE_RECORD,
2 NAME,
3 FIRST CHARACTER (10) VARYING,
3 MIDDLE_INITIAL CHARACTER ( 1) ,
3 LAST CHARACTER (15) VARYING,
2 ID_NUMBER FIXED DECIMAL(9),
2 SALARY FIXED DECIMAL ( 7 , 2) ;
26
The number in front of each component is a level number. The
structure as a whole is at level one. The members of the
level-one structure are the level-two components; those of
the level- two components are the level-three components, etc.
It is possible to write structure declarations using nonconse-
cutive level numbers, but there is always an equivalent
structure using consecutive ones. In any event, the logical
levels are always consecutively numbered.
The organization of a structured value is just like that
of a structured variable. A structured value contains a number
of components, and can be treated as a single object. For
example, two structures can be added just as two arrays can be
added. There are no structure constants.
The elements of a structure are referred to using
qualified names, although abbreviated versions are permissible.
The fully-qualified names of the elements of the structure
given above are:
EMPLOYEE_RECORD (itself a structure)
EMPLOYEE_RECORD.NAME (itself a structure)
EMPLOYEE_RECORD. NAME. FIRST
EMPLOYEE_RECORD . NAME . MIDDLE_INITIAL
EMPLOYEE_RECORD . NAME . LAST
EMPLOYEE_RECORD . I D_NUMBER
EMPLOYEE_RECORD . SALARY
Abbreviated versions of qualified names are obtained by leaving
out any of the component identifiers other than the last one.
27
These abbreviated forms can be used as long as the result is
not ambiguous, i.e., as long as it cannot refer to more than
one object.
Arrays of structures and structures of arrays are possible,
and can be nested to any depth. An example of such a nested
structure is given by:
DECLARE
1 CAR (30) ,
2 COUNTRY_OF_ORIGIN FIXED DECIMAL(3),
2 DEALERS (40) ,
3 CITY CHARACTER (20) VARYING,
3 STATE CHARACTER (2) ,
3 COMPANY CHARACTER (30) VARYING;
In such a nested entity, dimensionality is inherited. Thus
CITY is a two-dimensional array since one dimension is inherited
from CAR and the other from DEALERS. CITY(*,17) designates
the array composed of the elements
CAR(l). DEALERS. CITY (17)
CAR(2). DEALERS. CITY(17)
CAR (30) .DEALERS. CITY(17)
In writing a reference to an element of CITY or a similar
object, the subscripts can be written in any position as long
as they are in the right order. Thus CITY(*,17) could also
have been written as CAR(*) .CITY ( 17) or as DEALERS (*, 17) .CITY .
28
DECLARATIONS
A declaration associates a set of attributes with an
identifier. The attributes specify the characteristics of the
object denoted by the identifier, such as its data type.
Within an external procedure, a given identifier can be declared
more than once, since the identifier can be declared in differ-
ent blocks, or as a member of different structures, or both.
Nevertheless, each declaration of an identifier designates a
distinct object having its own attributes. The rules for name
resolution determine how an occurrence of an identifier is
resolved to the appropriate declaration (see "Blocks and Scopes"
below) .
Every occurrence of an identifier within an external
procedure must have a corresponding declaration within that
external procedure. Moreover, every declaration must have a
complete and consistent set of attributes. To satisfy these
principles, PL/I includes extensive conventions for defaulting
declarations, i.e., for creating declarations that were not
written in the program and for deducing attributes of incomplete
declarations.
Manifest, Explicit, Contextual, and Implicit Declarations
An identifier that is declared in a DECLARE-statement
written by the programmer is said to be man-i ;'-r ^ '' u declared.
* The term "manifest", though convenient, is not standard
usage .
29
An identifier that is manifestly declared, or that appears in
a parameter list, or appears as a statement name, is said to
be ex-plioitly declared. An identifier appearing in a parameter
list may, but need not, be manifestly declared; an identifier
appearing in a statement-name must not be manifestly declared.
If an identifier appears in a procedure and no explicit
declaration exists for that identifier, then a default declara-
tion is created if possible. The default declaration is placed
in the outermost block of the external procedure. If the
identifier is used in such a way as to suggest what its
attributes should be, it is said to be contextually declared;
otherwise it is implicitly declared. For instance, if the
identifier OUT has not been manifestly declared and the state-
ment
PUT FILE (OUT) LIST(VALX,VALY) ;
appears, then OUT will be contextually declared with the attri-
butes FILE and CONSTANT. Similarly, if the identifier EXP has
not been manifestly declared and the statement
Y = EXP(A**2) ;
appears, EXP will be contextually declared with the attribute
BUILTIN. A program is in error if it induces conflicting
contextual declarations.
An implicit declaration is created for an identifier
if it is declared neither explicitly nor contextually. In
this case, the created declaration initially has no attributes,
and all the attributes are added by default later on.
30
Declarations of Statement-Names
The appearance of an identifier as a statement-name
completely determines its attributes. The data type of the
identifier is determined by the kind of statement named, as
well as whether or not the identifier is subscripted. If the
named statement is an ENTRY-statement or a PROCEDURE-statement ,
the identifier acquires the attributes ENTRY and CONSTANT; if
the named statement is a FORMAT- statement the identifier
acquires the attributes FORMAT and CONSTANT; and if it is any
other statement the identifier acquires the attributes
LABEL and CONSTANT. Moreover, if the statement-name is sub-
scripted, then the identifier acquires an appropriate
DIMENSION-attribute .
The implied declaration of an entry constant cannot be
fully constructed until the types of its parameters are known.
When the parameter types are known, the entry constant can be
characterized completely. At that point the parameter speci-
fications and the RETURNS-attribute , if any, are added to the
declaration. For instance, the statements
FN: PROCEDURE (A, B) RETURNS (CHARACTER ( 12) VARYING) ;
DECLARE A CHARACTER (*) ;
DECLARE B POINTER;
END FN;
lead to the derived declaration
DECLARE FN ENTRY (CHARACTER (*), POINTER)
RETURNS (CHARACTER (12) VARYING) CONSTANT;
31
Attribute Consistency and Completeness
The diagram of Figure 2 can be used to determine whether
a set of attributes is consistent and complete, except for a
few peculiar cases. In this diagram, the following conventions
are followed:
1. Double lines indicate the definition of a term.
2. Rectangles indicate specific attributes. If a rectangle
is dashed, it indicates an optional attribute.
3. Ovals indicate sets of attributes.
4. Horizontal lines indicate sets for which all (nondashed)
elements must be present) .
5. Vertical lines indicate sets from which one alternative
must be chosen.
A set of attributes is complete and consistent if it can be
constructed from this diagram; it is consistent if it is a
subset of a complete and consistent set.
The interpretation of this diagram is illustrated by the
attribute set
INTERNAL VARIABLE AUTOMATIC ALIGNED POINTER INITIAL
Starting from the node "consistent-set", a scope and a declara-
tion-type must both be chosen. The scope can be either INTERNAL
or EXTERNAL; INTERNAL is chosen. The declaration-type can be
a variable, a named-constant, etc.; it is chosen to be a
variable. In that case, the attribute VARIABLE must be present,
as well as a storage-type and a data-declaration. The
storage-type is chosen to be a storage-class; the storage-class
32
o'
O
03
-P
0)
0)
4-1
XI
■H
>-l
-U
-P
<
4-)
C
0)
4-1
•rH
C
o
U
c
rO
0)
4-1
Q)
i-H
a
e
3
•H
Dd
Q
W
U
H
a
?=:
U
w
H
?:
D3
^
Q
<
s
Aa
Z
z
o
M
M
Eh
e-i
U
1— 1
M
Q
D
2
CQ
O
U
v^
0)
a
o
u
\^ S
J
J
<
<
z
•~r
cc:
a
w
CjJ
H
H
Z
X
M
DJ
4J
dJ
I
4-J
C
(U
4J
in
c
v8y
01
0}
1T3
rH
O
I
<u
en
(0
U
O
4-1
^§
CJ
Q
H
W
Eh
U
M
Q
W
1-^
S
H
.^_
W
O
o
<
<
K
H
H
CQ
H
P
m
2
<
O
U
en I
2
2
<
Eh
U^
2
O
U
I
c
4-1
0)
c
o
u
I
n
e
o
4-1
a
•H
u
0)
0)
I
(1)
vi/
>H
a:
U
Eh
1 1
2
M
U
iu
ir<
CQ
O
^
tu
I ^1
E-«
H
X
-a
(U
c
•H
■P
C
o
u
(1)
•H
TT
a
Di
>i
P
4-1
H
W
1
•^-
U
«
(0
D
M
4J
K
1-1
(0
Eh
TJ
Ol
^
O
H
M
u
w
/^
r^
cr>
(T3
^
w
hJ
<JJ .
J
cu
T!
^""
<
—
X
O
w
o
e
Pi
u
o
M
tn
W
2
H
Q
O
■H
4J
a
u
en
0)
I
rC
-P
CO
^
■N
(
/^
rj
^"
1 —
—
1
z
X
J
o
M
w
a
) 5H
rf:
2
>-l
0)
J
h-l
< :£;
?:
M
a
•u
<
tx,
a
J ^<
i..
M
^"" ^ ^
<
u
■; W
^
c
) z
C)
Di
>
b
Di
u
u
1 M
Q
<
>
z
o
o
1
z
^^
N
J
J
—
y
/
>.
I
_.
FT r-
w
t
c
D Q
H
Eh
oi
t
n W
<
fc
H
U
p
f
0 = X
^
O
b— . 1-
H
^
H
j:
2 H
J
C
u
U
frn
Cn
<
•X,
u
H
^
J '-'
'-"^n
<^
>i
-P
-P
1
ex
>1
4J /
1 (
^ (
K,
f
"N
rH
0
rH
M
Q.
/^
C
•H
4-)
(T3
P
c
J-
a)
O
•H
to
P
O _
•H - ,
5
i-l
4-
c
C
Qj
4-)
3
-P
P
4->
"■"
>i
(0
a.
(0
H
in
c
h
+j
—
+J -
- e
4-)
M
1
■u
:3
0
u
c
3
^ V
-J
o
(0
0
0
0
I "^J
^
v^
I u ]
34
S'
•Si
I ^1
Q
W
w
(1)
c
■H
4-)
c
o
u
(N
0)
z
o
M
El
O
Oi
I
a:
2
a.
4J
I
c
o
■H
4J
(13
-U
D
a
E
O
o
_<
^
Eh '
0.1
ui lui
o o
I I I
1-1
c
CJ
a:
u
H
H
W
►-
7:
—
W
M
Uh
O
u.
(X
o
o
o
+J
(fl
u
o
J v^
35
I 2
W
I ^
' 2
O
I °^
>
I
0
4-1
a,
•H
u
10
0)
'O
I
\^
■^■
<
M
H
H
u
2
u
W -
- a
3
M
a
Q
w
CO
J
w
Eh
u <
Q
Cl,
-p
4-1
cu
0)
If)
tn
T1
u
(1)
o
S-l
u
-p
(U
w
u
w ^-^
is chosen to be AUTOMATIC. The data-description may, but
need not, contain DIMENSION; in this case it does not contain
DIMENSION. The data-description must also contain an alignment,
chosen to be ALIGNED, and either a data-type with optional
INITIAL, or STRUCTURE. In this case, INITIAL is present
(the actual initial values are ignored) . The data-type is
chosen to be noncomputational . Of the alternatives for non-
computational-type, locator is selected; of the alternatives
for locator, POINTER is selected. Thus the entire set of
attributes is shown to be complete and consistent. (The order
in which the attributes are given is immaterial.)
Not all consistency violations are shown up by reference
to this diagram. For example, the combination
AUTOMATIC EXTERNAL
and the combination
STATIC LABEL INITIAL
are both invalid, but pass the test of the diagram. Specific
auxiliary rules are needed in order to disallow these and
similar cases. The requirement for completeness is relaxed
for the case of file constants, i.e., declarations with the
attributes FILE and CONSTANT; although the file-description-set
must be consistent, it need not be complete.
Certain attributes either permit or require subspecif ica-
tion; these are listed in Table 3. For example, the BASED
attribute permits, but does not require, the subspecif ication
of an auxiliary pointer (see "Based Storage" below) , while the
36
Table 3. Attributes Requiring or Permitting Subspecif ication
A. Subspecif ication Required
Attribute
CHARACTER
BIT
AREA
DIMENSION
RETURNS
PRECISION
PICTURE
GENERIC
ENTRY with CONSTANT
POSITION
DEFINED
LIKE
Required
Specification
length
length
size of area
dimensionality
type of returned value
number-of-digits, possible scale-factor
picture specification
generic specification
parameter types
position count
base variable
likened variable
B. Subspecif ication Permitted
Attribute
Permitted
Subspecif ication
BASED
OFFSET
ENTRY with VARIABLE
basing pointer
area
parameter types
37
CHARACTER attribute requires the subspecif ication of a string
length.
The ENTRY-attribute and the RETURNS-attribute themselves
contain attribute sets as subspecif ications . These attribute
sets are known as descriptor's , and must also be complete and
consistent. The descriptors in the ENTRY-attribute describe
the parameters expected by a procedure, while the descriptors
in a RETURNS-attribute describe the value returned by a proce-
*
dure. To be complete and consistent, a descriptor must be
derivable from the "data-type" node in the diagram of Figure 2,
but may optionally contain the MEMBER-attribute . The descriptor
for a structure as a whole looks like the declaration of a
structure variable with the identifiers left off, e.g.,
DECLARE GENFUNC ENTRY (
1 (*) ,
2 FIXED BINARY,
2 CHARACTER (30) VARYING );
Standard and User Defined Defaults
The PL/I defaulting rules specify how an incomplete but
consistent set of attributes is to be completed. Although
there is a standard set of defaulting rules, the DEFAULT-
statement can be used to override them. The defaulting rules,
whether standard or user-defined, consist of a predicate and
a default attribute set. The predicate indicates a test to
be applied to the attributes already present in the declaration
* Since procedures can accept structures as arguments and
can return structures as values, the specification of a
single parameter or of a returned value can contain more
than one descriptor.
38
(or descriptor) , while the default attribute set indicates
additional attributes to be supplied provided that they are
consistent with the ones already present. Inconsistency, in
this case, is not an error; it simply means that the default
is not applied.
The principal standard defaulting rules are:
1. Add the attributes FIXED, REAL, and BINARY. For instance,
*
if FIXED alone is present, REAL and BINARY are added.
2. If the arithmetic attributes are present but no precision
has been specified, add an implementation-defined precision
whose subspecif ication depends on the arithmetic attri-
butes already present.
3. If CHARACTER or BIT is present, assume NONVARYING.
4. If CHARACTER or BIT is present but no length has been
specified, assume a length of 1 . If AREA is present but
no area size has been specified, assume an implementation-
defined value for the area size. If POSITION is present
but no count has been specified, assume a count of 1.
5. If neither VARIABLE nor CONSTANT has been specified,
assume VARIABLE unless ENTRY or FILE is present. If ENTRY
or FILE is present by itself, assume CONSTANT. If ENTRY
or FILE is present along with other attributes, the default
depends on what those other attributes are.
6. If FILE and CONSTANT, or ENTRY and CONSTANT, or CONDITION
is present, assume that the scope is EXTERNAL; in all
other cases assume that it is INTERNAL.
' This rule differs from the one used in the well-known IBM
implementation of PL/I.
39
7. If EXTERNAL is present, assume that the storage class is
STATIC; in all other cases assume that it is AUTOMATIC.
8. If CHARACTER or BIT is present, assume UNALIGNED;
otherwise assume ALIGNED.
As a consequence of the defaulting rules, an implicitly declared
identifier, which starts with an empty attribute set, will
acquire the attribute set
REAL FIXED BINARY PRECISION (d , 0 ) VARIABLE INTERNAL
AUTOMATIC ALIGNED
Here d, is the implementation-defined default number-of-digits
for the attribute combination FIXED BINARY. Similar constants
d.2,d^, and d are specified for FIXED DECIMAL, FLOAT BINARY,
and FLOAT DECIMAL.
The DEFAULT-statement contains a predicate and one or more
default attribute sets. Declarations are completed by applying
the user-supplied DEFAULT-statements in the order that they
appear in the program, and then applying the system default
rules. Unlike DECLARE-statements , DEFAULT-statements are order-
dependent. A default attribute set is added if and only if
the appropriate predicate is satisfied and none of the elements
in the set leads to a conflict. The predicate can include both
attributes and identifier ranges, indicated by the keyword
RANGE. The predicate is formed as a boolean combination of
attributes and ranges. RANGE (*) is satisfied by any identifier,
but not by a descriptor. RANGE (al;a2) is satisfied by any
* This rule does not apply to structures. The alignment of a
structure, if given explicitly, is passed down to all element-
ary members of the structure unless a conflicting alignment
is given at a lower level.
40
identifier starting with a letter between al and a2 inclusive,
while RANGE (init) is satisfied by any identifier starting
with init. For instance,
DEFAULT (RANGE (AB) I FLOAT & ""BINARY) COMPLEX STATIC;
applied to the declaration
DECLARE AB35 FIXED;
yields
DECLARE AB3 5 FIXED COMPLEX STATIC;
and applied to the declaration
DECLARE XYZ FLOAT DECIMAL;
yields
DECLARE XYZ FLOAT DECIMAL COMPLEX STATIC;
However, it has no effect when applied to
DECLARE AB35 REAL FIXED;
since COMPLEX conflicts with REAL. It also has no effect when
applied to
DECLARE XYZ FLOAT BINARY;
since the predicate is not satisfied.
The LIKE-Attribute
The LIKE-attribute can be used to copy part of a structure
declaration into another declaration. It is useful when a
program uses many similarly organized structures. For instance,
if a program includes the declarations
41
DECLARE
1 ASSEMBLY BASED,
2 NEXT_PART POINTER,
2 FIRST_COMPONENT POINTER,
2 DESCRIPTION,
3 PART_NUMBER PICTURE •X(5)9',
3 COST FIXED DECIMAL ( 6 , 2) ;
DECLARE 1 GROUP (20) STATIC LIKE ASSEMBLY;
then the second declaration is equivalent to
DECLARE
1 GROUP (20) STATIC,
2 NEXT_PART POINTER,
2 FIRST_COMPONENT POINTER,
2 DESCRIPTION,
3 PART_NUMBER PICTURE 'X(5)9*,
3 COST FIXED DECIMAL ( 6 , 2 ) ;
The LIKE-attribute causes copying of members only; attributes
at the level of the LIKE-attribute are not copied. In this
example, a reference to GROUP (20) .NEXT_PART requires that the
qualifying identifier GROUP be included, since otherwise the
reference would be ambiguous. This behavior is a general
property of structures declared using the LIKE-attribute.
42
EXPRESSIONS, TYPE CONVERSION, AND ASSIGNMENT
The kinds of expressions acceptable in PL/I are similar
to those found in most higher-level languages. These are:
literal constants
references to variables and named constants
parenthesized expressions
function calls
prefix expressions
infix expressions
references to builtin functions
The only kinds of literal constants recognized are arithmetic
constants and string constants. Other constants are obtained
either as named constants, e.g., statement-names, or as the
results of builtin functions, e.g., NULL. References to
variables may have subscripts, pointer qualifications (see
below), and name qualifications. An example of a reference
with all three is
PT -> X(I, J) . B
Although the pointer-qualifier symbol -> looks like an operator
it is not treated as one (see "Based Storage" below) .
Parentheses are used within expressions in three ways:
to designate subscripts of arrays, to designate arguments of
functions, and to group components of expressions containing
operators. When an expression containing operators, e.g., +
and *, is enclosed in parentheses, it is treated as a single
43
entity -- this is the usual convention in mathematical nota-
tion. When a reference to a variable is enclosed in paren-
theses, it is then treated as a general expression rather
than as a variable. This rule only makes a difference in
the context of a procedure call.
A function call has two parts: the reference to the
function and the argument list. The function reference need
not have the form of a single identifier, since functions
can return entry values, can be subscripted, and can be
members of structures. Function calls, subscripted references,
and references to builtin functions all have the same syntactic
form, so a knowledge of the relevant declarations is necessary
in order to distinguish them. An example of a rather complex
function call is
A,B(3) (I, 'NEXT' ) (X)
In this case, A,B(3) is an element of an array of structures
(or a structure of arrays) containing an entry value. That
entry value designates a procedure that expects two arguments
— in this case, I and the string constant 'NEXT' — and
itself returns an entry value. The entry value obtained from
this second procedure is then applied to the argument X. A
function expecting no arguments is called by using an empty
argument list. Thus
NEXT_SUIT( )
would call the procedure NEXT_SUIT with no arguments.
If E is a procedure that itself returns an entry value, then
44
F(E)
indicates that F is to be called with an argument consisting
of the entry value associated with E, while
F(E( ))
indicates that F is to be called with an argument obtained
by calling E as a function of no arguments. This convention
is somewhat different from the one used in Algol 60 and many
of its descendants.
Function calls are discussed in more detail under
"Procedures, Scopes and Environments" below.
Prefix and Infix Expressions
A prefix expression consists of an expression preceded by
a prefix operator, while an infix expression consists of two
expressions with an infix operator between them. When an
expression contains a string of operators, the meaning is
determined by the precedences of the operators. Those opera-
tors with highest precedence are applied first, then those of
next highest precedence, etc. The infix operators, grouped
by precedence from high to low, are:
* /
+
II
&
I
Table 4 summarizes the meanings of the operators.
45
Table 4. PL/I Operators and Their Meanings
Infix Operators
*
/
+
<
>
< =
>=
-<
-•>
exponentiation
multiplication
division
addition
subtraction
concatenation
equal
not equal
less than
greater than
less than or equal
greater than or equal
not less
not greater
Prefix Operators
minus
plus
not
46
When a sequence of adjacent operators, all of the same prece-
dence, appears, the operators are applied from left to right
except in the case of **, which is applied from right to left.
Thus
A*B/C**D**E
is interpreted as
(A * B) / (C ** (D ** E) )
Prefix operators are always applied first unless they conflict
with the ** operator; in that case the ** is applied first,
so that
is interpreted as
even though
- A ** 3
- (A ** 3)
- A * 3
is interpreted as
(- A) * 3
There are five arithmetic operators in PL/I:
+ addition
subtraction
* multiplication
/ division
** exponentiation
In order to apply any of the first four, the operands must
first be converted to a common base, scale, and mode according
to the following rules:
47
Base; binary if either operand binary, otherwise decimal
Scale: float if either operand float, otherwise fixed
Mode; complex if either operand complex, otherwise real
The operands need not have the same precision, however. The
rules for the results of these operations assume that there
is a maximum value N for the number-of-digits of the result.
The "precision rules" then give the number-of-digits (and,
for the case of fixed, the scale-factor) of the result.
They are arranged so that digits on the right are never
thrown away except in the case of division, where it cannot
be avoided. If the operands are float, then the number-of-
digits of the result is the maximum of the nuinbers-of-digits
of the operands. Otherwise, assume that the two operands,
which are necessarily fixed, have precision (p,q) and (r,s)
respectively. The result precision (m,n) for the four opera-
tions is given by:
+,- m = min(N, max(p-q,r-s) + max(q,s) + 1)
n = max ( q , s )
* m = min (N,p+r+l)
n = q+s
/ m = N
n = N-p+q-s
Should the result value exceed the capacity of the result
precision, the FIXEDOVERFLOW- condition is raised, indicating
an error (see "ON-Units and ON-Statements" below) .
The situation with regard to exponentiation is somewhat
more complicated. Suppose that the formula x ** y is being
48
computed. If either x or y is float, the result is float,
and the result precision is that of p. If x is real and
fixed, and y is a small integer constant, then the result
is also real and fixed, and the precision of the result
is given by:
m= (p + 1) *y-l
n = q * y
In any other case where both x and y are real, the result is
real. If either x or y is complex, the result is complex.
In certain cases, such as x real and fixed with a negative
value and y not an integer, an error is indicated.
The comparison operators are:
= equal
"■= not equal
< less than
> greater than
<= less than or equal to
>= greater than or equal to
"■< not less than
"■> not greater than
The operators <= and ""> are equivalent, as are the operators
>= and "■< ; "■> and ""< are included mainly for intellectual
compatibility with COBOL, which uses the phrases NOT GREATER
and NOT LESS. All of the comparison operators return a
one-bit value: 'I'B if the comparison is satisfied, and 'O'B
if it is not. The equality and inequality comparisons can be
applied to any type of data, although for most of the non-
printable types both operands must have the same type. (The
49
only exception is that pointers can be compared to offsets.)
The comparison operators that test for inequality cannot
be applied to complex arithmetic data or to nonprintable data,
but they can be applied to real arithmetic data, to pictured
data, and to strings. The meanings of these operators
applied to arithmetic data are the usual ones; a numeric
pictured datum is treated as the numeric value that it repre-
sents. When character strings of unequal length are compared,
the shorter one is filled on the right with blanks to bring it
to the same length as the longer one. The ordered comparisons
are then done left to right on the basis of the implementation-
defined collating sequence, which defines an order for the
individual characters. For instance, the letters are ordered
alphabetically and the digits numerically. Two character
strings compare equal if they are identical after the shorter
one has been blank-filled on the right. The ability to
perform ordered comparisons of character strings is particu-
larly useful in applications that involve sorting names.
Similar rules apply to bit strings: If two bit strings are
of unequal length, the shorter one is filled on the right with
zero-bits prior to the comparison. If the strings differ, the
comparison is done bit-by-bit from the left with the rule that
a one-bit is greater than a zero-bit.
If the two operands of a comparision have different types,
they are converted to a common type. If one has an arithmetic
type, the other is converted to an arithmetic type. If neither
has an arithmetic type, but one has a character type and the
50
other has a bit type, the bit- type operand is converted to
character.
The concatenation operator, "| |", is used to put two
strings together. For instance, the value of
•AVER' I I 'AGE'
is the string 'AVERAGE'. If one operand is a bit string and
the other is a character string, the bit string is converted
to a character string (as it is for comparisons) .
There are also three logical operators:
& and
I
"" not
"&" and "I" are infix operators, while """' is a prefix operator,
The operands of these operators are expected to be bit strings,
so that if they have any other type, they are conerted to bit
strings .
Builtin Functions
PL/I includes a large variety of builtin functions. A list
of these, together with a brief explanation of what each one
does, is given in Table 5. Some of the more important builtin
functions will now be described.
The first group of builtin functions deals with strings.
The descriptions of the functions will be given for character
strings, but the definitions for bit strings are analogous.
The builtin function LENGTH (x) returns as its value the
51
actual length of the character string x. It is useful in
two contexts: determining the length of a string passed as
a parameter, and determining the current length of a varying
string. In both of these cases, the declaration of the
string does not provide enough information to determine the
length.
The builtin function SUBSTR (x, y_, z) is used to extract a
portion of a string. x is the string, y is the position of
the first character to be extracted, and z^ is the number of
characters to be extracted. If z^ is omitted, all of the string
starting with the character at position y. is extracted. The
null string is a possible result. SUBSTR can also be used
on the left side of an assignment. For example, given the
statements
DECLARE CHARS CH ARACTER ( 8 ) ;
CHARS = ' TOM JONES ' ;
SUBSTR(CHAR8,2,5) = 'IMHA';
the resulting value of CHARS is 'TIM HANES ' . A builtin function
used on the left side of an assignment in this way is called
a pseudovariable.
The builtin function INDEX (x,^;) finds the first position
within the string x where the string ^ occurs. If y. does not
occur at all within x, the value of INDEX is 0. For example:
value of INDEX (' SYNCOPATION • , 'COP') = 4
value of INDEX (' SYNCOPATION ' , 'COPE') = 0
The builtin function VERIFY (x,^) finds the first character in
52
Table 5. SUMMARY OF THE PL/I BUILTIN FUNCTIONS
In this table, descriptions of the various PL/I builtin
functions are given. These descriptions are intended to
indicate the intent of each function, and in some cases the
principal restrictions on their arguments. In the descriptions
of the functions, square brackets are used to indicate optional
arguments .
1. ABS(.r) - the absolute value of x. If x is complex,
the value is its modulus.
2. ACOS(a') - the arc cosine of x. x must not be complex.
3. KDT){x,y ,p , \q]) - the sum of x and u with precision
{p,q) or (r.O) if the result is fixed, and with precision
(p) if the result is float.
4. ADDR(.t) - a pointer to the generation of x.
5. AFTER (sa,ca) - the portion of the string sa that follows
the first occurrence of ca within sa. If aa does not
occur within sa , the value is the null string.
6. ALLOCATION (x) - the number of generations of the controlled
variable x that currently exist.
7. ASIN(j:) - the arc sine of x. x must not be complex.
8. ATAN (j/ [,.r] ) - the arc tangent of y/x if x is given,
and of 1/ otherwise.
9. ATAND(!/[ , J-] ) - the arc tangent in degrees of y/x if x is
given, and of y otherwise. x and y must be real.
10. ATANH{x) - the hyperbolic arc tangent of x.
53
Table 5. Continued
11. BEFORE (sa,ea) - the portion of the string sa that precedes
the first occurrence of aa within sa . If ca does not
occur within sa, the value is the null string.
12. BINARY (x [ ,p [ ,(7] ] ) - the result of converting x to binary,
with precision determined by p and q if one or both is given.
13. BIT(j;, [le]) - the result of converting x to bit, with
length le if le is given.
14. BOOL{x ,y , oa) - the boolean function of x and y whose truth
table is specified by the four-bit value ca.
15. CEIL(a;) - the least integer greater than or equal to x.
X must be complex.
16. CHARACTER (sa, [Z-e ] ) - the result of converting sa to charac-
ter, with length le if le is given.
17. COLLATE ( ) - the implementation-defined collating sequence,
as a character string.
18. COMPLEX (x, 2/) - the complex number x + i y.
19. CONJG(x) - the complex conjugate of x.
20. COPY(sa,Ze) - the string consisting of le copies of sa
concatenated together.
21. COS(a;) - the cosine of x.
22. COSD(a:) - the cosine of x, with x given in degrees.
X must not be complex.
23. COSH(x) - the hyperbolic cosine of x.
24. DATE ( ) - the current date, in the form yymmdd, where
yy is the year, mm is the month, and dd the day.
54
25. DECAT {sa, ca , pa) - a portion of the string sa. sa is
partitioned into three pieces by the first occurrence
of aa. The three-bit string pa specifies which of the
three pieces (before oa, aa itself, after ca) are to be
concatenated to form the value of DECAT.
26. DECIMAL {x [ ,p [ ,^] ] ) - the result of converting x to deci-
mal, with precision determined by p and q if one or both
is given.
27. DIMENSION (x, n) - the number of elements in the n-th
dimension of the array x, defined as HBOUND(x,n)-
LBOUND(x,n) + 1.
28. DIVIDE (x,i/ ,p [,q]) - the quotient of x and y with
precision {p,q) or (p,0) if the result is fixed, and
with precision (p) if it is float.
29. DOT{x,y [ ,p I ,q] ]) - the dot product of x and u, with
precision determined by p and q if one or both is given.
30. EMPTY ( ) - the empty area-value.
31. ERF(x) - the statistical error function of x.
32. ERFC(x) - the complanent of the statistical error func-
tion of X.
33. EVERY(x) - the value 'I'B if every bit in x is a one-bit,
and 'O'B otherwise. For this purpose, all scalar-elements
of X are converted to bit,
34. EXP(x) - the exponential function of x.
35. FIXED (x, p[,q]) - the result of converting .r to fixed,
with precision determined by p and, if it is given, ,.
36. FLOAT (x,p) - the result of converting x to float with
precision (p) .
55
37. FLOOR(x) - the greatest integer less than or equal to x.
X must be real.
38. HBOUND(a:,n) - the upper bound of the n-th dimension of
the array x.
39. HIGH(Ze) - a string of le copies of the highest character
in the collating sequence.
40. IMAG(a:) - the imaginary part of the complex number x.
41. INDEX (sa,c?a) - the position within the string sa of the
first occurrence of the string aa. The value is 0 if oa
does not occur within sa.
42. LBOUND(x,n) - the lower bound of the n-th subscript of
the array x.
43. LENGTH (sa) - the length of the string sa .
44. LINENO(/n) - the current line number (within a page) of
the print file fn.
45. LOG(x) - the natural logarithm of x.
46. LOGIO (x) - the logarithm to the base 10 of x. x must
not be complex.
47. L0G2(x) - the logarithm to the base 2 of x. x must not
be complex.
48. LOW(Ze) - a string of le copies of the lowest character
in the collating sequence.
49. MAX(x ,x„ , . . . ,x ) - the largest of the numerical values
of the X. . The x. must not be complex.
50. MIN (x- ,x_, . . . ,x ) - the smallest of the numerical values
of the X. . The x. must not be complex.
11
51. MOD(x,iy) - the value of x modulo y. x and y must not be
complex.
56
52. MULTIPLY(x, z/,p [,(^1 ) - the product of x and y with preci-
sion {p,q) or (p,0) if the result is fixed, and with
precision (p) if it is float.
53. NULL{ ) - the null pointer.
54. OFFSET ipt ,ar) - the result of converting the pointer
pt to an offset within the area ar .
55. ONCHAR ( ) - the leftmost erroneous character within the
current ONSOURCE -value . When the conversion-condition
is raised, the current ONSOURCE -value is set to the string
whose conversion was being attempted, and it retains this
value during the execution of the associated on-unit.
56. ONCODE ( ) - an implementation-defined integer indicating
the nature of the on-condition associated with the current
on-unit .
57. ONFIELD ( ) - the contents of an erroneous field encountered
during data-direct input, causing the name condition to be
raised.
58. ONFILE( ) - the name of the file being processed when an
input-output condition was raised.
59. ONKEY ( ) - the name of an erroneous key that caused the
KEY-condition to be raised during record input-output.
60. ONLOC ( ) - the name of the procedure entry-point active
when the current on-unit was raised.
61. ONSOURCE ( ) - the current onsource-value . When the CONVER-
SION -condition is raised, the current ONSOURCE -value is
set to the string whose conversion was being attempted,
62. PAGENO(/>i) - the number of the current page within the
print file fn.
57
63. POINTER (0/6,0^)- the result of converting the offset
ofe within the area ar to a pointer.
64. PRECISION (x,p [,(?] ) - the result of converting x to
precision (p/C?) or (p,0) if x is fixed, and to precision
(p) if X is float.
65. PROD(x) - the product of all the elements of the array x.
66. RE7U1. (x) - the real part of the complex number x.
67. REVERSE (sa) - the bits or characters of the string sa
taken in reverse order.
68. ROUND (x,n) - the result of rounding up the numerical value
of x. If a; is fixed, the result has a scale-factor of n;
otherwise the result has a number-of-digits of n.
n must be an integer constant.
69. SIGN(x) - the value +1, 0, or -1 according to whether x is
positive , zero or negative .
70. SIN(x) - the sine of x.
71. SIND(x) - the sine of x, with x given in degrees.
X must not be complex.
72. SINH(x) - the hyperbolic sine of x.
73. SOME(x) - the value 'I'B if at least one bit in x is a
one-bit, and 'O'B otherwise. For this purpose, all
scalar-elements of x are converted to bit.
74. SQRT(x) - the square root of x.
75. STRING (sa) - the result of concatenating together the
scalar-elements of sa after converting them to bit.
76. SUBSTR(sa, st [ , Ze] ) - the substring of sa consisting of le
characters or bits of sa beginning with the st-th one.
58
Table 5. Continued
If le is omitted, the substring consists of the
characters or bits from the si-th one to the last.
77. SUBTRACT(x, i/,p [ ,(7] ) - the difference of x and y with
precision {^ ,q) or (p,0) if the result is fixed, and
with precision (p) if it is float.
78. SUM(x) - the sum of all the elements of the array x.
79. TAN(a:) - the tangent of x.
80. TAND(x) - the tangent of x, with x given in degrees.
X must not be complex.
81. TANH (x) - the hyperbolic tangent of x.
82. TIME ( ) - the current time, in the form hhmmss where
hh gives the hour, mm gives the minute, and ss...s gives
the second carried to an implementation-defined number of
fractional decimal places.
83. TRANSLATE (sa,ra [, pa] ) - the result of replacing, within
sa, each character of pa by the corresponding character
of ra . If pa is omitted, it is taken to be the collating
sequence .
84. TRUNC(x) - the result of truncating x to the nearest integer
in the direction of zero. x must not be complex.
85. UNSPEC(x) - the internal representation of x, as a bit string
86. VALID(f3a) - the value 'I'B if the current value of the
pictured variable ea conforms to the picture and 'O'B
otherwise .
87. VERIFY (8a,r!a) - the position within the strin--! • • of the
first character or bit of SA that does not ir within oa .
aa thus behaves as a set rather than a sequence. If all
characters or bits of sa occur within , the value of VERIFY
is 0. 53
the string x that is not a character of the string y. If all
the characters of x appear in y , then the value of VERIFY is 0.
For example:
value of VERIFY ( 'CABDRIVER' , 'ABCDE') = 5
value of VERIFY ( 'CEDE' , 'ABCDE' ) = 0
The builtin function REVERSE (x) reverses its argument, so for
example:
value of REVERSE ( 'GALLOP') = ' POLLAG '
This function is useful in right-to-left scanning; the string
to be scanned is reversed and then scanned left to right. The
builtin function COLLATE ( ) (the "( )" indicates that COLLATE
is a function of no arguments) has as its value the implemen-
tation-defined collating sequence, i.e., the string consisting
of all acceptable characters ordered from least to greatest.
The builtin function COPY(x,n) creates n copies of the string x.
For example,
value of COPY( 'CHA' , 3) = 'CHACHACHA'
The arithmetic builtin functions enable the user to control
the attributes of arithmetic results. These fall into two
groups. First, there are builtin functions ADD, SUBTRACT,
MULTIPLY, and DIVIDE that behave like the corresponding infix
operators, except that the precision of the result is explicitly
specified. For example,
MULTIPLY (Ml , M2 , 5 , 3 )
produces a result whose precision is (5,3), and whose remaining
attributes are determined by the attributes of Ml and M2 .
60
Second, there are builtin functions FIXED, FLOAT, BINARY, and
DECIMAL that convert their argument to the specified attribute
with the specified precision. For instance,
DECIMAL(M1,4,2)
converts Ml to fixed decimal with precision (4,2) ; the mode of
the result is the mode of Ml.
The conversion builtin functions FIXED, FLOAT, BINARY, and
DECIMAL can be used not only to convert among arithmetic types
but also to convert from the string types. The rules for the
conversion are discussed under "Type Conversion" below. There
are further conversion functions REAL(j-) which converts x to
real type (and for a complex number, takes its real part) ;
IMAG(x), which takes the imaginary part of the complex number
X (and yields 0 if x is not a complex number) ; and COMPLEX (x, z/ ) ,
which converts x and z/ to a common real type and then forms
the complex number x + ii/ . For conversion to string types, the
builtin functions CHARACTER (x, n) and BIT(a:,n) can be used.
CHARACTER ( X, ^) first converts x to character type and then
adjusts the length of the result to n either by truncating on
the right or by filling on the right with blanks. BIT behaves
similarly, either truncating or filling with zero-bits. The
second argument of either of these functions may be omitted,
in which case no truncation or filling is done.
The mathematical builtin functions included in PL/I are
listed in Table 6. With the exceptions indicated, they can
accept arguments of any arithmetic type, including c- x
types. For some mathematical functions there e than one
61
possible range for the result value, and the choice of principal
value is specified in the table.
The function ATAN (arctangent) can accept either one or
two arguments. The two-argument version is useful in converting
rectangular coordinates to polar coordinates. If the rectangular
coordinates are given by the pair {x,y), then ATAN{x,y) gives
the corresponding polar angle in the range from -it to +tt . Since
the value of the tangent function repeats every tt/2 radians,
the sign of y is needed to determine the correct value.
A number of the builtin functions fall into no particular
category. The builtin function SUM(j:) accepts an array as
argument, and returns as value the sum of all the elements of
the array. The builtin function PROD, for "product", behaves
similarly. The builtin function DOT[x,y) expects its arguments
to be one-dimensional arrays both having the same bounds;
it takes the mathematical dot product of x and y. The builtin
function BOOLix ,y , z) takes as arguments two bit strings x and
y of arbitrary length, and a third bit string z of length 4.
z determines a boolean function that is applied to x and y .
If z is the sequence b^b^b ^b . , then the function is defined
by;
bit of X bit of y result
0 0 2?^
0 1 ^2
1 0 2»3
1 lb.
62
Table 6. Mathematical Builtin Functions
PL/I Mathematical
Name Description
ABS absolute value
ACON arc cosine
ASIN arc sine
ATAN arc tangent (one argument)
ATANp arc tangent of quotient
(two arguments)
ATAND, arc tangent in degrees
(one argument)
ATAND- arc tangent of quotient
in degrees (2 arguments)
ATANH hyperbolic arctangent
COS cosine
COSD cosine in degrees
COSH hyperbolic cosine
ERF error function
ERFC complement of error func-
tion
EXP exponential
LOG natural logarithm
L0G2 base 2 logarithm
LOGIO base 10 logarithm
SIN sine
SIND sine in degrees
SINH hyperbolic sine
SQRT square root
TAN tangent
TAND tangent in degrees
TANH hyperbolic tangent
Complex Constraints on
Arguments? Result R
(Principal Value)
yes
R 5^ 0
no
0 < R < TT
no
-tt/2 < R < TT/2
yes
-7T/2 < R < tt/2
no
no
yes
yes
no
yes
no
no
yes
yes
no
no
yes
no
yes
yes
yes
no
yes
(real argument)
-TT < Re (R) < TT
(complex argument)
-90 < R < 90
-180 < R < 180
-TT < Im(R) < TT
Re(R) > 0 or
Re(R)=0 and Im(R) >0
63
The builtin function VALID can be used to check the
validity of pictured data, i.e., to ensure that the value
stored in a pictured variable fits the description given by
the picture. VALID (j:) returns 'I'B if the pictured variable
X contains a valid value, and 'O'B otherwise. Invalid values
can arise since an arbitrary character string can be read
into or assigned to a pictured varible, and ordinarily no
validity check is made at the time of reading or assignment.
The builtin functions EVERY and SOME are useful in
testing properties of aggregates. EVERY (x) returns 'I'B if
its argument (after conversion to bit type, if necessary)
consists entirely of one-bits, and 'O'B otherwise. SOME (x) ,
on the other hand, returns 'I'B if its argument contains at
least one one-bit, and 'O'B otherwise. For example, if A is
an array of arithmetic type, then the expression A > 0 will
be an array with a one-bit in each position i where A{i) > 0.
Therefore EVERY (A>0) will return 'I'B if all elements of A
are greater than 0, while SOME(A>0) will return 'I'B if at
least one element of A is greater than 0.
Type Conversion
In PL/I it is possible to convert from any printable
type to any other, although for certain values the conversion
may be illegal. Conversions may be invoked either explicitly,
using builtin functions such as FLOAT or CHARACTER, or
implicitly in contexts such as operands of operators or
arguments of functions. For instance, the concatenation
64
operator requires that its operands be strings of the same type
(bit or character) , so that the operands must be converted
appropriately — even if they are of arithmetic type. The
text of a procedure defines the types of its parameters, and
if the arguments of a procedure do not already have the expect-
ed types, they too must be converted. In fact, PL/I provides
implicit conversions in almost every context where conversion
is possible.
The conversions among arithmetic types generally follow
the principle of preserving the meaning of the converted
value. For example, the result of converting the fixed value
7.3 to complex float decimal with precision (8) is
. 73000000E+01+0I . In conversion to a fixed type when digits
must be dropped, the result value is obtained by truncating
towards 0, although in certain unusual cases an implementation
may produce a slightly different result. When converting
from real to complex an imaginary part of 0 is added, while
when converting from complex to real the imaginary part is
dropped .
The conversion between bit and character is straightfor-
ward; zero-bits correspond to the character "0", and one-bits
correspond to the character "1". A character string to be
converted to bit type must consist entirely of these two
characters, or an error is signalled — specifically, the
CONVERSlON-condition . It is possible for the programmer to
modify the converted value so as to correct the error (see
"Categorization of the ON-Conditions" below) .
65
The most complicated conversions are those between the
string types and the arithmetic types. A character string is
converted to a number by treating the string as the represen-
tation of a number. Thus, given the statements
DECLARE NUMV FIXED DECIMAL (5 , 2 ) ;
NUMV = ']6]62.13E1]6' ;
the string ']6]62 .13E1\6' is converted first to the float value
that it represents, and then to the fixed decimal value 21.30.
The blanks surrounding the number are always permissible.
An all-blank value converts to zero. If the character string
does not represent a valid number, then the CONVERSION-condi-
tion is signalled. As in the case of conversion from character
to bit, it is possible for the programmer to correct the error.
Conversion from a number to a character string yields,
in effect, the result of printing the number. Ordinarily that
result includes leading blanks. For instance, the effect of
DECLARE NUMV FIXED DECIMAL (4);
DECLARE CONV_RESULT CHARACTER (20 ) VARYING;
NUMV = 17;
CONV_RESULT = NUMV;
is to assign the string ']^)6]6]6]611 ' to NUMV. In most cases the
length of the resulting string is the number-of-digits after
conversion (if necessary) to fixed decimal, plus three.
Three spare positions are needed in order to accommodate a
possible sign, a possible decimal point, and a possible leading
zero.
66
Conversion from an arithmetic value to a bit string is
accomplished by first converting the arithmetic value to real
fixed binary and then converting the integer part of the
value to the corresponding bit string. For instance, convert-
ing the value 12.6 to a bit string yields 'OOllOO'B, with an
intermediate conversion from fixed decimal with precision
(3,1) to fixed binary with precision (10,4), (The rules for
obtaining the intermediate precision are somewhat complicated,
but it can be seen that two digits to the left of the decimal
point may require as many as six nonzero bits to represent
their value.) Conversion from a bit string to an arithmetic
type is accomplished by treating the bit string as a binary
number, and then converting from that number to the desired
type.
It is also possible to convert from pointer to offset,
or vice versa, provided that an area is given. Thus if AR
is an area and P is a pointer, the expression OFFSET(P,AR)
gives the result of converting P to an offset relative to A.
Similarly, if OFS is an offset, POINTER (OFS , A) gives the
result of converting OFS to a pointer relative to A. A pointer
can be declared with an area-reference, as in
DECLARE AR2 AREA;
DECLARE 02 OFFSET ( AR2);
In this case, 02 can implicitly be converted to a pointer,
and the .;:^.i AR2 is used in the conversion.
67
Promotion
The PL/I operators, and many of the builtin functions
also, can be applied to aggregates as well as to scalars. When
two aggregates of the same organization, i.e., two structures
with equal numbers of components or two arrays of the same
dimensionality, are used as the operands of an operator, the
result also has that organization, and the result is formed
by combining corresponding components of the operands.
For example, in
DECLARE A (3, 4) FIXED BINARY
DECLARE B{3,4) FIXED BINARY
DECLARE C(3,4) FIXED BINARY
A = B + C;
the assignment to A is accomplished by adding B(l,l) to C{1,1),
B(l,2) to C(l,2), etc., to form a new array of sums with
dimensionality (3,4). The array of sums is then copied into A.
In certain peculiar cases the temporary array containing the
sura is actually needed, and it does not suffice simply to add
the elements of B and C one by one and place the result directly
in A.
It is also possible to combine scalars with structures,
scalars with arrays, and structures with arrays. However, it
is not possible to combine structures having different numbers
of members, or arrays having different dimensionalities. A
scalar is combined with a structure by promoting it to a
similar structure, all of whose members have the same value as
the scalar. Similarly, a scalar is combined with an array by
68
promoting the scalar to a similar array. The case of combin-
ing a structure with an array is more complicated; first the
original structure is promoted to an array of structures, and
later each element of the original array is promoted to a
structure. A simple instance of promotion is given by the
expression A+1, where A is an array of arithmetic type. The
value of this expression is obtained by creating an array of
I's, having the same dimensionality as A, and then adding
this new array to A, element by element. The effect is just
the same as adding 1 to each element of A.
The Assignment-Statement
The assignment-statement contains a left side, which is
a list of targets, and a right side, which is an expression.
Each of the targets designates a location capable of receiving
a value. The statement is executed by evaluating the expres-
sion and then storing its value, after appropriate conversion,
into the location designated by each target, in order from
left to right. For instance, the assignment-statement
A, B(I) , C=l;
causes 1 to be stored into A, B(I), and C. The targets of
an assignment-statement may be variables or pseudo-variables.
For instance, the assignment-statement
SUBSTR(TEXT,I,LEN) = WORD;
stores the value of WORD into the indicated substring of TEXT
(after adjusting the size of the value to be LEN) . Similarly,
69
IMAG(Z) = SIN(X) ;
causes the imaginary part of the (necessarily) complex vari-
able Z to be set to the value of SIN(X) , while the real part
of Z is left undisturbed.
Since the type of the value obtained from the right side
of an assignment-statement may disagree with the type of a
target, promotion or conversion, or both, may be necessary.
If the target is a scalar, then the usual rules for scalar
conversion are applied; the type of the target defines the
type to which the value must be converted. If the target is
a structure or array, then the value must be promoted to the
type of that structure or array, by replicating elements as
necessary. Following the promotion, element-by-element scalar
conversion may be necessary. For instance, in the example
DECLARE HVAROO) FLOAT BINARY;
HVAR =0;
the scalar value 0 is promoted to an array of 30 fixed zeros.
Each of these is then converted to an appropriate float zero,
*
and assigned to the corresponding element of the array.
A variation on the assignment-statement, called
by-name assignment , can be used to assign elements from one
structure to another according to the names of the elements
rather than according to their positions in the structure.
For instance, given the declarations
* In actual practice, the conversion is usually done before
rather than after the promotion. The result is the same.
70
DECLARE
1 RED,
2 BLUE,
2 GREEN,
3 ORANGE,
3 WHITE,
2 BLACK ,
2 GRAY ;
DECLARE
1 VIOLET,
2 BLACK ,
2 WHITE,
2 GREEN,
3 ORANGE ,
3 VIOLET,
2 TAN
2 BLUE;
the effect of the assignment-statement
RED = VIOLET, BY NAME ;
is to perform theindividual assignments
RED. BLUE = VIOLET, BLUE;
RED. GREEN. ORANGE = VIOLET . GREEN .ORANGE;
RED. BLACK = VIOLET .BLACK ;
Those members not in common between the two structures are
ignored. By-name assignment can be extended to accommodate
expressions that involve structures.
71
STORAGE TYPES
PL/I provides a variety of ways to manage the storage
of variables. Each variable has a storage type, which can
be either parameter , defined, or a storage class. The parameter
and defined storage types indicate that the variable is an
alias, i.e., an alternate name, for storage that has already
been obtained by other means. The storage classes provide
different ways of allocating and freeing storage; the storage
classes are static, automatic , controlled , and based. The
storage type of a variable is determined by its declaration
after all defaulting of declarations has been done; in most
cases the default is the automatic type. The storage used
to hold the value of a variable is called a generation . A
generation can exist even though it is not currently associated
with any variable.
Static Storage
The static storage class is the simplest one. When a
variable is declared to be static, its generation is allocated
at the start of program execution and remains allocated
throughout program execution. When a static variable is
declared within a procedure, the values of that variable are
kept from one call of the procedure to the next. Even in the
case of a recursive procedure, there is just one copy of the
variable, and that copy is available at all levels of recur-
sion. Static storage is much like the standard form of storage
in FORTRAN.
72
Automatic Storage
Storage for an automatic variable is allocated on entrance
to the block where the variable is declared, and freed on exit
from that block. Whenever the block is entered, a fresh gener-
ation is obtained for the variable. In practice it sometimes
happens that values of automatic variables are retained from
one block entrance to the next, but this behavior is not any-
thing that the programmer can rely upon. When an automatic
variable is declared within a recursive procedure, a new
generation is created for each level of recursion, and remains
associated with the variable at that recursion level until
the recursion level is terminated. Automatic storage resembles
the ordinary local storage of Algol .
Controlled Storage
Controlled storage is explicitly allocated and freed by
the programmer using the ALLOCATE-statement and the FREE-state-
ment. Each time the variable is allocated, a new generation
for it is created and placed on a pushdown stack; each time the
variable is freed, the generation at the top of the stack is
destroyed. There is one such stack for each controlled variable,
and the current value of the variable is always obtained from
the generation at the top of the stack. In other words, the
generations follow a last-in-first-out rule.
The values of string lengths and array dimensions in the
declaration of a controlled variable can be given by expressions.
73
The expressions are evaluated when a new generation is allocated,
and so the different generations need not all have the same
sizes. For example, suppose that we are given the statements:
DECLARE N FIXED;
DECLARE CONTV CHARACTER (N) CONTROLLED;
N = 5;
ALLOCATE CONTV;
CONTV = 'FIRST' ;
N = 7;
ALLOCATE CONTV;
CONTV = 'SECOND' ;
If CONTV has not been previuosly allocated, these statements
will create a stack consisting of two generations. The genera-
tion at the top of the stack will have length 7 and value
'SECONDjzJ' (the assignment adds a blank on the right) , while
the other generation will have length 5 and value 'FIRST'.
Thus the current value of CONTV will be 'SECONDfc^'. If the
statements
FREE CONTV;
PUT LIST (CONTV) ;
are executed, then the generation at the top of the stack will
be destroyed and CONTV will refer to the first generation.
Consequently the PUT-statement will cause FIRST to be printed.
Based Storage
Based variables are useful in creating linked data struc-
tures, and also have applications in record input-output. A
based variable does not have any storage of its own; instead,
the declaration acts as a template and describes a generation
of storage. In order to use the variable to refer to a
74
particular generation of storage, a pointer to that generation
must also be provided. The pointer and the based variable,
taken together, constitute a based reference . In many cases,
the pointer is given implicitly rather than explicitly.
An example of a declaration of a based variable and a
pointer is
DECLARE
1 ARRAY_ELT BASED,
2 ARRAY (10) FLOAT,
2 NEXT_ELT POINTER;
DECLARE AP POINTER;
ARRAY_ELT describes a generation of storage, namely, a struc-
ture containing a float array and a pointer. The based
reference
A - > ARRAY ( 4 )
designates a particular element within the ARRAY_ELT structure
pointed at by the pointer AP , and if AP does not point at such
a structure the reference is invalid. PL/I does not provide
any mechanism for checking that a pointer is indeed pointing
at a generation of the correct type, and so it is entirely the
programmer's responsibility. The errors that result when a
pointer points at an object of the wrong type can often be
extremely difficult to track down.
The statement
ALLOCATE ARRAY_ELT SET(AP);
causes a generation matching the type of ARRAy_ELT to be
created and also causes the pointer AP to point at that genera-
tion. Thus, after this ALLOCATE-statement has been executed,
75
a reference to AP -> ARFIAY(4) will be valid. If subsequently
the statement
FREE AP -> ARRAY_ELT;
is executed (and the value of AP has not been changed in the
meantime) , the generation pointed at by AP will be destroyed,
and subsequent references to that generation will be meaning-
less .
In this example, the structure includes not only the
array but also a pointer. That pointer can be used to form
a list of arrays, each one pointing to its successor. An
element is added to the head of the list by allocating it
and setting its NEXT_ELT component to the previous list head.
Similarly, the head of the list is deleted by setting the
new list head to the NEXT_ELT component of the old list head
and then freeing the old list head. One of the main uses
of based variables and pointers in PL/I is constructing lists
such as this one. In order to end a list, a special null
pointer is needed, and that pointer is provided by the NULL
builtin function.
It is convenient to have to write a pointer with every
based reference. Therefore it is possible to declare an
implicit pointer in the declaration of a based variable, e.g.,
DECLARE BFIX BASED (BFP) FIXED;
DECLARE BFP POINTER;
A reference to BFIX by itself is taken to mean BFP -> BFIX.
Moreover, the statement
76
ALLOCATE BFIX;
is equivalent to
ALLOCATE BFIX SET(BFP);
and the statement
FREE BFIX;
is equivalent to
FREE BFP -> BFIX;
The template given by a based variable can be applied to
storage of types other than based. In order to obtain a pointer
to a generation, the ADDR builtin function is used. ADDR(y)
yields a pointer to the generation specified by y . As an
example, the statements
DECLARE BCOMP FLOAT COMPLEX BASED;
DECLARE SCOMP FLOAT COMPLEX STATIC;
ADDR(SCOMP) -> BCOMP = 2E0 + 3E0I;
cause the static variable SCOMP to be set to the value 2E0+3E0I.
The Refer-Option
The string lengths and array bounds of a based variable
can be specified by expressions as well as by constants. For
example, the declaration
DECLARE ECS CHARACTER (M) BASED;
indicates that the length of BCS is given by the current value
of M. When BCS is allocated, the generation that is created
will have a length given by the current value of M, and when
reference is made to BCS, the value of M must agree with the
77
length of the string in the generation referred to. If a
number of generations, all corresponding to BCS , exist, it
may be difficult to ensure that the current value of M is
correct, since the generations may have different string
lengths. In order to deal with this difficulty, PL/I allows
the string length to be specified along with the string
itself; both the string and the length are stored in a single
structure, sometimes called a set f- defining structure. For
instance, the structure
DECLARE
1 STRING_STRUC BASED (STP),
2 LEN FIXED,
2 NEXT POINTER,
2 STRING CHARACTER ( LEN 1 REFER(LEN));
DECLARE STP POINTER;
could be used to create a list of strings, each having a
different length. When one of these structures is allocated,
the length of the string is obtained as the current value of
LENl, and at the same time the current value of LENl is auto-
matically stored within the LEN component of the newly created
generation. When one of these structures is referenced, the
length of STRING is obtained from the LEN component of that
structure. Both LEN and LENl are needed, for the following
reason. Were LEN used without the so-called refer-option,
the allocation size would be taken from STP->LEN prior to the
allocation, which would be either undefined or the string
length of a previously allocated generation. On the other
hand, were LENl used, it would then be necessary to reset it
to LEN before referencing STRING, since otherwise the length
78
of STRING would not be correct.
A based variable may contain any number of ref er-options .
These can be used to specify upper or lower array bounds as
well as string lengths and area sizes.
Lef t-to-Right Correspondence
It is often necessary to create data structures in which
the elements do not all have the same type, as in the following
example:
DECLARE
1 FLOAT_ELEMENT BASED (ELPTR) ,
2 ELTYPE FIXED, /* 1 FOR FLOAT */
2 NEXT POINTER,
2 VALUE FLOAT;
DECLARE
1 FIXED_ELEMENT BASED (ELPTR) ,
2 ELTYPE FIXED, /* 2 FOR FIXED */
2 NEXT POINTER,
2 VALUE FIXED;
DECLARE
1 CHAR_ELEMENT BASED (ELPTR) ,
2 ELTYPE FIXED, /* 3 FOR CHARACTER */
2 NEXT POINTER,
2 VALUE CHARACTER (24 ) ;
DECLARE ELPTR POINTER;
A list can be formed containing elements of all three kinds,
storing a type code in ELTYPE in order to distinguish among
them. In order to reference an element, it is necessary to
specify either FLOAT_ELEMENT, FIXEDELEMENT, or CHAR_ELEMENT
even before the type of the element is known, since a refer-
ence to ELTYPE by itself is syntactically ambiguous. Therefore
under certain conditions PL/I allows a reference to a component
79
of a based structure even when the variable in the reference
does not agree with the generation being referenced. The
primary condition is that the generation and the variable must
agree up to that component, although there are further detailed
requirements that are beyond the scope of this article. Based
references satisfying this constraint are said to be in
left-to-right aorrespondence , since they agree reading from
left to right. Thus it is permissible to use
FLOAT_ELEMENT.ELTYPE to refer to, and therefore to test, the
type code stored in any one of the three kinds of elements.
Even if ELPTR is pointing at a generation having the type of
CHAR_ELEMENT, the ELTYPE component of that generation can be
referenced using FLOAT_ELEMENT. ELTYPE. Since ELTYPE is the
first component of each element, the elements necessarily
agree up to that component. Moreover, the NEXT components of
the three kinds of elements can be referenced interchangeably
since in each kind of element NEXT has type pointer and is
preceded by an element having type fixed (with the remaining
attributes defaulted identically in all cases) .
Allocation in Areas
A based variable can be allocated in a specified area,
as in the following example:
DECLARE BV FIXED BASED (P);
DECLARE A AREA(20 0);
ALLOCATE BV IN (A) ;
Since BV has been allocated in A, the OFFSET builtin function
80
can be used to convert P into an offset relative to A, as
given by
OFFSET(P,A)
The allocation can assign a value directly to an offset
variable, as in the example
DECLARE OFS OFFSET (A) ;
DECLARE BVl BASED (OFS);
ALLOCATE BVl IN (A) ;
Since BVl is based on OFS, the offset of BVl relative to A is
assigned to OFS when the ALLOCATE-statement is executed.
Parameter Storage
A variable acquires the parameter storage type by virtue
of its appearance in a parameter list of either a PROCEDURE-
statement or an ENTRY-statement . The PARAMETER attribute can,
but need not, be declared for a parameter; it is invalid to
use that attribute for any other kind of variable. A parameter
describes a generation of storage passed as an argument to the
procedure that declares the parameter. Thus, allocation and
freeing of the parameter is the responsibility of the procedure's
caller. Since a parameter is allocated before the procedure
declaring it is entered, the procedure itself cannot specify
an initial value for the parameter. See "Arguments and
Parameters" below for further information about parameters.
81
Defined Storage
The defined storage type, like the parameter storage
type, is an alias. The declaration of a defined variable
specifies a base item, which is a portion (or possibly all)
of some other variable. The defined variable provides another
way of referencing part or all of the storage occupied by the
base item. The base item can be part of a variable having any
storage type other than defined or based, and so circular
defining is excluded.
There are three kinds of defining: simple-defining ,
isub-de fining, and overlay-defining . The sort of defining that
is in effect is determined by the relation between the defined
variable and the base variable. Since defined variables are
aliases, they are not allocated nor freed, nor are initial
values specified for them.
An example of simple-defining is
DECLARE A (5, 8) FIXED;
DECLARE ADEF(2:4) DEFINED (A (1 ,*)) ;
ADEF is defined to consist of the elements A(l,2), A(l,3), and
A(l,4). For simple-defining to be in effect, the attributes
of the defined variable must agree with those of the base item,
except that the array bounds of the defined variable may be more
restrictive than the corresponding bounds of the base item. A
major use of simple-defining is to specify portions of arrays
that are to be passed as arguments to procedures.
82
Isub-def ining is in effect when the base item contains
special subscripts, known as isubs. These subscripts have
the form ISUB, 2SUB, etc. An example of isub-def ining is
DECLARE A (10, 10) FIXED;
DECLARE ADEF(9,8) DEFINED (A (lSUB+1 , 2SUB+2) ) ;
A reference to an element of ADEF is translated into a refer-
ence to an element of A by substituting the first subscript
for ISUB and the second subscript for 2SUB. For instance,
ADEF(8,6) refers to A(9,8). The defined array need not have
the same dimensionality as the base item. For example, in
DECLARE B{30,30) FLOAT;
DECLARE BDIAGOO) DEFINED (B ( ISUB, ISUB) ) ;
the one-dimensional array BDIAG consists of the diagonal
elements of the array B, while in
DECLARE C(15) POINTER;
DECLARE C2(5,3) DEFINED (C { 3* ( lSUB-1) +2SUB) ) ;
the two-dimensional array C2 is defined onto the one-dimensional
array C.
Overlay-defining is used in order to apply different
descriptions to strings. For the purposes of overlay-defining,
character data and pictures are together considered as
aharaater-alass data, while bit strings are considered as
bit-claee data. An excimple of overlay-defining is:
83
DECLARE CS CHARACTER ( 30 ) ;
DECLARE 0DEF1{3) CHARACTER (5) POSITION (10) DEFINED(CS);
DECLARE
1 0DEF2 DEFINED (CS) ,
2 OCSl CHARACTER (14 ) ,
2 0CS2 PICTURE '$$$V.$$'; /* 6 CHARACTER POSITIONS*/
The relationship between CS and ODEFl, and between CS and 0DEF2,
is illustrated in Figure 3. The POSITION attribute in the
declaration of ODEFl indicates that the character sequence
comprising ODEFl starts at character 10 of CS . ODEF(l)
consists of characters 10-14 of CS , ODEFl (2) of characters
15-19 of CS, and ODEFl (3) of characters 20-24 of CS . The
treatment of 0DEF2 is similar. In overlay-defining both the
defined variable and the base item must consist entirely of
unaligned data (see "Alignment" below) of the same class, but
a string can be overlaid onto an array as well as the other
way round .
Alignment
The declaration of a variable can specify an alignment ,
either ALIGNED or UNALIGNED. An aligned variable is stored so
as to favor speed of access over space; typically, storage for
an aligned variable is placed at a word boundary or other
natural demarcation for the machine at hand. An unaligned
variable is stored so as to favor space over speed of access,
and is arranged in storage so as to minimize unused space.
The default alignment for nonvarying strings and for pictures
is unaligned; for everything else it is aligned.
84
cs
1[ 2| 3| 4| 5| 6| 7| 8| 9p.0[Iip.2(I3(14|15[16[I7tI8tL9|20|21[22[23t24^5|26l27l28|29|30
ODEFl(l) 0DEF1(2) ODEFlO)
(a) Overlaying ODEFl onto CS
CS
1| 2| 3| 4| 5| 6| 7| 8| 9tL0tLliL2tL3p.4|I5[L6tL7tl8ll9|20l21^2|23|24|25|26^7l28l29|30
OCSl 0CS2
(b) Overlaying 0DEF2 onto CS
Figure 3. Example of Overlay-Defining
85
In most situations, the alignment of a variable has no
effect on its behavior. The exception is that aggregates
composed of unaligned strings and pictures are stored with
all their components adjacent, i.e., as a sequence of adjacent
characters (or bits, in the case of unaligned bit strings).
The sequence can then be used as a base item for overlay-
defining.
Initialization
It is possible to specify an initialization for a vari-
able, as long as its storage type is not an alias, i.e., is
neither parameter nor defined. The initialization is specified
using the INITIAL-attribute. For example, in the declaration
DECLARE A (40) FIXED INITIAL ( (40 ) 0) ;
the array A is initialized to all zeroes. The initialization
can be specified by a single item, by a repeated item, or by
a repeated list, which can itself contain items of these types.
Nesting to any depth is permitted. Thus
DECLARE B(20) FIXED INITIAL ( 2 , 3 , ( 5) 4 , ( 3) (-1 , -2 ) ) ;
causes the first 13 elements of B to be initialized to the
sequence
2 3 4 4 4 4 4-1-2-1-2-1-2
For a multidimensional array, initializations are performed
with the last subscript varying most rapidly. Thus
DECLARE C(3,2) FIXED INITIAL (1 , 2 , 3 , 4 , 5 , 6 ) ;
86
causes the initializations
C(l,l) = 1 C(l,2) - 2
C(2,l) = 3 C(2,2) = 4
C(3,l) = 5 C(3,2) - 6
Initialization always takes place at the time of alloca-
tion. Thus, for static variables, the initialization is
performed at the start of program execution. For automatic
variables, it is performed at each entrance to the declaring
block. For controlled and based variables, it is performed
when an ALLOCATE-statement for the variable is executed, and
is applied to the newly allocated generation. Parameters
and defined variables cannot be initialized with the
INITIAL-attribute .
87
PROCEDURES, SCOPES, AND ENVIRONMENTS
Textually, a procedure is a body of code, delimited by
a PROCEDURE-statement at the beginning and an END-statement
at the end. Associated with the procedure are one or more
entry points , each of which provides a way of invoking some
portion of the code contained within the procedure. The entry
points are defined by the PROCEDURE-statement, as well as by
any ENTRY-statements that appear within the procedure. The
characteristics of an entry point include its name, the
niomber and types of its parameters, and the type of its
returned value, if any. Each entry point, in turn, defines
an entry constant.
A procedure is called either by means of a function
reference within an expression or by means of a subroutine
reference within a CALL-statement . The procedure call itself
consists of an entry-valued reference and an argument list,
possibly empty. For instance, the procedure call
F(X+3, ' INVALID' ) has an entry-valued reference F and an
argument list consisting of the two arguments X+3 and
'INVALID'. The value of F must be an entry point of the
procedure being called. Usually F is just the name of the
procedure, but F could also be, for instance, an entry vari-
able. An empty argument list for a function reference must be
indicated explicitly by ( ) . If an entry point returns a
value, then it m.ust be called by a function reference; other-
wise it must be called by a subroutine reference.
88
An example of a procedure definition is:
PI:
P2: PROCEDURE (QVAL, SIZE) RETURNS (FLOAT BINARY) ;
DECLARE QVAL FLOAT BINARY;
DECLARE SIZE FIXED DECIMAL(4);
DECLARE J FIXED BINARY;
DECLARE TOTAL FLOAT BINARY INITIAL (0);
DO J = 1 TO SIZE;
TOTAL = TOTAL + F(QVAL,J);
END;
RETURN (TOTAL) ;
P3: ENTRY (RVAL, RES, SIZE) ;
DECLARE RVAL FLOAT BINARY;
DO J = 1 TO SIZE;
RES = RES + F(RVAL,J);
END;
RETURN
END PI
This procedure has three entry points. PI and P2 are synonymous
(but do not compare equal) , and are entry constants designating
the entry point at the PROCEDURE-statement . Since that entry
point returns a value (with attributes FLOAT BINARY) , PI and
P2 can only be called as function references, i.e., as compon-
ents of an expression. P3 is the entry constant naming the
entry point starting at the ENTRY-statement . It does not return
a value, and so P3 can only be invoked from a CALL-statement ,
e.g., by
CALL P3(A(2) ,B(2) ,22) ;
P2 has two parameters, namely, QVAL and SIZE, while P3 has three
parameters, namely, RVAL, RES, and SIZE. As this example shows,
the entry points need not have the same parameters, and if any
parameters are in common, they need not appear in the same
position. It is invalid to reference a parameter not associated
89
with the entry point actually used to enter a procedure. For
instance, it is invalid to reference RVAL if the procedure is
entered through PI or P2.
The RETURN-Statement
The RETURN- statement is used to end execution of a proce-
dure. It may have either the form
RETURN (expr) ;
or the form
RETURN ;
If the procedure is called by a function reference, then the
RETURN-statement must contain an expression. Conversely, if
the procedure is called by a subroutine reference, then the
RETURN-statement must not contain an expression.
When a RETURN-statement containing an expression is
executed, the expression is evaluated. The value of the
expression is then taken as the value of the function reference
that called the procedure. If necessary, the value of the
expression is converted to the type specified in the RETURNS-
clause of the entry point where the procedure was entered.
A procedure can return an aggregate as well as a scalar.
Moreover, the returned type may have asterisks in its speci-
fication , e.g.,
RETURNS (CHARACTER ( * ) ) ;
for an entry point that returns a character string of arbitrary
length.
90
Execution of a RETURN-statement not containing an expres-
sion ends execution of the procedure and causes control to
return to the point of call. The END-statement of a procedure
is treated as having an implicit RETURN-statement just in front
of it, so that if control flows to the END-statement, execution
of the procedure is terminated. It is an error to allow control
to flow to the END-statement of a procedure that was called
as a function reference.
Arguments and Parameters
An entry point of a procedure can have a sequence of
parameters associated with it. A call on the entry point must
include a corresponding sequence of arguments, which act as
inputs to the procedure. If
(1) the argument is a reference to a variable (possibly
subscripted or name-qualified) , and
(2) the attributes of the argument agree with those
of the parameter,
then the parameter becomes an alias for the argument, and
assignments to the parameter affect the argument. In all
other cases, the argument is considered to be a dummy. That
is, when the call is made, a generation of storage -- the
dummy -- is set aside for the argument, and the value of the
argument is copied into that generation. If the type of the
argument disagrees with the type of the parameter, the argu-
ment is converted to the parameter type and the converted value
is assigned to the dummy. The parameter is then an alias for
91
the dummy, and after the call is completed the dummy is
discarded. Thus, assignments to a parameter that corresponds
to a dummy argxament have no effect at the point of call.
Constants and expressions are always passed as dummy arguments.
PL/I uses the call-by-reference model of argument
transmission, i.e., the location of the argument is passed to
the procedure. The conventions for argument transmission are
shown by the following example:
CALLER :
CALLEE ;
PROCEDURE;
DECLARE X FIXED DECIMAL(5);
DECLARE Y FLOAT DECIMAL(7);
CALL CALLEE (X); /* X IS SET TO 12 BY THE CALL */
CALL CALLEE (Y); /* DUMMY CREATED, SO Y IS UNCHANGED
CALL CALLEE (24962) ; /* DUMMY CREATED */
CALL CALLEE ( (X.) ) ; /* DUMMY CREATED SINCE (X) IS
AN EXPRESSSION */
CALL CALLEE (X+1 4 ) ; /* DUMMY CREATED HERE, TOO */
END CALLER;
PROCEDURE (P) ;
DECLARE P FIXED DECIMAL(5); /* P IS THE PARAMETER*/
P = 12;
END CALLEE;
Array sizes, string lengths, and area sizes of parameters
must be given either by constant-valued expressions or by
asterisks. An asterisk size is used when the size of the
92
corresponding argument is unknown, or varies from one call to
another. Thus a parameter declared as CHARACTER (*) will match
an argument declared as CHARACTER ( e ) , where e is any expression-
However, such a parameter will not match an argument declared
as CHARACTER (e) VARYING.
Options
Implementation-defined information can be attached either
to a PROCEDURE-statement or to the declaration of an entry
constant by means of the OPTIONS-attribute . A particularly
common option (but not a universal one) is illustrated by
PROCEDURE OPTIONS (MAIN) ;
where the MAIN option indicates that execution of the program
is to start with this procedure. In general, the information
given in an OPTIONS-attribute affects the manner in which the
procedure is compiled.
When a PL/I procedure references a procedure written in a
different programming language, the OPTIONS-attribute can be
used to specify the language of that foreign procedure so that
appropriate calling sequences can be compiled. For instance,
DECLARE PRIMFN ENTRY (FLOAT) RETURNS (FLOAT)
OPTIONS (FORTRAN) ;
would describe a procedure written in Fortran to be called from
a procedure written in PL/I .
93
Recursion
A PL/I procedure is permitted to call itself, either
directly or indirectly. A procedure that calls itself is
said to be reaursive , and the RECURSIVE option must be
specified on the PROCEDURE-statement of such a procedure.
An example of a recursive procedure is one that counts the
number of nodes in a binary tree. Each node is represented
as a based structure, and contains a value, a left son, and
a right son. Each son is either itself a pointer to a binary
tree, or null. The procedure in PL/I is:
COUNTNODES :
PROCEDURE (NODEPTR) RECURSIVE RETURNS (FIXED) ;
DECLARE (LCOUNT,RCOUNT) FIXED INITIAL(O);
DECLARE
1 NODE BASED (NODEPTR) ,
2 LEFT_SON POINTER,
2 RIGHT_SON POINTER,
2 VALUE FIXED;
DECLARE NODEPTR POINTER;
DECLARE NULL BUILTIN;
IF LEFT_SON "= NULL THEN
LCOUNT = COUNTNODES (LEFT_SON) ;
IF RIGHT_SON "= NULL THEN
RCOUNT = COUNTNODES (RIGHT_SON) ;
RETURN (LCOUNT+RCOUNT+1) ;
END COUNTNODES;
The procedure is given a pointer to a binary tree as an
argument, and it returns the number of nodes in the tree as
its value. Recursiveness is a property of a procedure rather
than of its entry points, so that even if the recursive call
is on a different entry point, the procedure is still
considered to be recursive.
94
The GENERIC-Attribute
Often it is useful to create a family of entry points that
perform a similar function but that expect somewhat different
arguments, and to assign a single name to the family. The
GENERIC-attribute allows a single name, known as a genevia
function , to be used for such a family of entry points; the
choice of entry points then depends on the nature of the argu-
ments. The GENERIC-attribute specifies a list of entry-point
names, and associates a sequence of generalized descriptors
with each name. A reference to the generic function is
translated into a reference to the first entry point whose
descriptors, as given by the GENERIC-attribute, match the
arguments of the generic function. An asterisk indicates a
descriptor that matches anything. The test for descriptor
matching is satisfied if the descriptor in the GENERIC-attribute
is contained in the attribute set of the argument; the attribute
set can contain attributes not in the descriptor. Only data
attributes can be tested in this way.
An example of a GENERIC-attribute is
DECLARE GF GENERIC (Gl WHEN (FIXED, FIXED) ,
G2 WHEN (FIXED, *) ,
G3 WHEN (*) ) ;
Using this declaration, and assuming the furtl^er declarations
DECLARE X FIXED BINARY;
DECLARE Y FLOAT DECIMAL;
the reference
95
GF(X,X+1) translates to G1(X,X+1)
The reference
GF(X,Y+1) translates to G2(X,Y+1)
since the expression Y+1 has data type float. The reference
GF(X) translates to G3(X)
since the first two descriptor sequences each require two
arguments .
Another application of the GENERIC-attribute is
illustrated by the declaration
DECLARE VARFN GENERIC (NFl WHEN (FLOAT (1 : 20 ) BINARY),
NF2 WHEN (FLOAT (21: 40) BINARY),
NFS WHEN (*) ) ;
In this case, the entry point represented by VARFN is selected
on the basis of the precision of the argument, which is assumed
to be float binary. If the argument has from 1 to 20 binary
digits, NFl is used; if the argument has from 21 to 40 binary
digits, NF2 is used; and if the argument has more than 40
binary digits, NF3 is used.
Blocks and Scopes
A block consists of a sequence of statements, starting
with a PROCEDURE-statement or a BEGIN-statement and extending
to the matching END-statement . Blocks of either kind can be
nested. The primary effect of the block structure of a program
is to define the scope of a name, i.e., the set of statements
from which the name can be referenced. A name declared in a
96
DECLARE- statement belongs to the innermost block containing
that DECLARE- statement. However, a name can also be declared
by virtue of its appearance as a parameter or as a statement-
name. A statement-name that names a PROCEDURE-statement , an
ENTRY-statement, or a BEGIN-statement belongs to the block
outside the one that contains that statement; any other
statement-name belongs to the block containing the statement
that it names. This rule is needed in order to allow
procedures to be called from the outside. A reference to a
name is resolved by searching the nest of blocks for a declar-
ation of the name, working from the inside out, and starting
with the statement containing the reference. Another way
of looking at it is that the scope of a name consists of the
block declaring the name and all contained blocks except
for those in which the scope is occluded by an inner declara-
tion of the same name.
An example illustrating the scope of names is given in
Figure 4. The parenthesized numbers are used to distinguish
different declarations of the same identifier. There is an
imaginary outer block used to hold the declarations of the
entry points of the external procedure (A and B in this case) ,
This block is needed since the entry points of a procedure
belong, not to the block of the procedure itself, but to the
next outer block. Since there is no such block for the
external procedure, one must be created.
97
Figure
4. Example Illustrating Scope
1
A:
PROCEDURE;
2
DECLARE X CHARACTER ( 1 ) ;
3
DECLARE B FIXED;
4
statement sequence 1
5
B:
ENTRY (Y) ;
6
statement sequence 2
7
C:
BEGIN ;
8
DECLARE W FIXED;
9
DECLARE Y PICTURE '(6)$';
10
statement sequence 3
11
D:
PROCEDURE;
12
DECLARE W FLOAT COMPLEX;
13
statement sequence 4
14
END D;
15
END C;
16
E:
PROCEDURE;
17
DECLARE W FLOAT;
18
Statement sequence 5
19
F:
B = 3;
20
G:
ENTRY ;
21
statement sequence 6
22
END E;
23
END A;
Statements belonging to
different blocks:
Names belonging to
different blocks:
outer
none
outer
A(l)
1,2,3,4,5,6,23
A(l)
C(7)
7,8,9,10,15
D(ll)
11,12,13,14
C(7)
E(16)
16,17,18,19,20,21,22
D(ll)
E(16)
A{1) , B(5)
X(2) ,B(3) ,Y(5) ,C(7) ,
E(16) ,G(20)
W(8) ,Y(9) ,D(11)
W(12)
W(17) ,F(19)
Statements in scopes of these names:
A(l)
X{2)
B(3)
B(5)
Y(5)
C(7)
W(8)
1-23
1-23
1-23
none
1-6,16-23
1-23
7-10,15
Y(9)
D(ll)
W(12)
E(16)
W(17)
F(19)
G(20)
7-15
7-15
11-14
1-23
16-22
16-22
1-23
98
Internal and External Scope
In most cases, declarations are defaulted to have internal
scope, meaning that the declaration designates an object
distinct from the objects designated by other declarations
of the same identifier. For instance, if the variable Q is
declared in three different blocks of a procedure with the
INTERNAL-attribute (possibly by default) , then each of these
blocks has its own distinct Q. However, several declarations
can be made to refer to the same object by giving them external
scope. External scope cannot be applied to just any declara-
tion; it is restricted to the static and controlled storage
classes, and to named constants. Identifiers declared to be
external must necessarily have the same attributes. As an
example, if the declaration
DECLARE A (14) STATIC EXTERNAL;
appears in two different blocks, then both declarations refer
to a single array. If an assignment is made to A(4) while
executing one of these blocks, then the change will be visible
in the other. Declarations of an external identifer can
appear both within a single external procedure and among several
external procedures. The default scope for manifestly declared
entry constants is external, since external procedures have
to be declared by the programmer while internal procedures are
automatically declared.
99
Entry Values and Environments
On account of the rules for scope of names in PL/I , a
procedure can refer to names in blocks surrounding the proce-
dure. Moreover, an entry point defines an entry value, and
that value can be assigned to an entry variable and subsequent-
ly invoked. Invocation of the entry point, in turn, requires
that references to outer-block names be resolved properly.
In order to achieve this effect, an entry value contains not
only a designation of an entry point but also an environment.
When the block surrounding the entry point is entered, the
environment of the entry point is defined. The entry value
corresponding to the entry point then consists of the entry
point itself together with a record of all names inherited
from outer blocks and the variables (or constants) that these
names denote. In the case of recursive procedures, the
environment implicitly designates not only a set of variables,
but also a recursion level. The following example illustrates
these concepts :
P : PROCEDURE ;
Q: PROCEDURE (R, LEVEL) RECURSIVE;
DECLARE R ENTRY; ■
DECLARE LEVEL FIXED;
IF LEVEL=10 THEN
CALL R( ) ;
ELSE IF LEVEL=6 THEN
CALL Q(S,7) ;
ELSE
CALL Q(R,LEVEL+1) ;
S: PROCEDURE;
PUT DATA (LEVEL) ;
STOP ;
END S;
END Q;
T: PROCEDURE;
END T;
CALL Q(T,1) ;
END P; ^QQ
The call on Q on the next- to-the-last line initiates a nest
of recursive calls. On each call, the value of LEVEL increases
by 1 . At the top level, the entry value T is passed as an
argument to Q; but since this entry value is never invoked,
it serves only as a place-holder. At the sixth level of
recursion, the entry value S is passed as part of the recur-
sive call on Q. The environment of this entry value consists
of the current set of outer-block variables -- outer, that is,
to S . In particular, since S is internal to Q, the current
value of LEVEL — 6 in thii. case -- is part of the environment
accompanying the entry constant S. On subsequent recursive
calls, the entry value is simply passed along (cf. the call
with parameters R and LEVEL+1) . When the recursion level
reaches 10, R is called. The value of R is the entry constant
S obtained at level 6, and so
LEVEL = 6
is printed out and the program halts.
The PL/I rules for block structure, scoping of names, and
environments are derived from Algol 60. In fact, it is possible
to transcribe the example above rather directly into Algol 60,
and the behavior in Algol 60 would be the same.
101
ON-UNITS AND ON-STATEMENTS
One of the more innovative aspects of PL/I is the facility
that it provides for handling exceptional conditions — the
so-called ON-conditions . An exceptional condition may arise
either as the result of an error, such as a subscript out of
range, or from an anticipated event, such as encountering
end-of-file while reading from a dataset or reaching the end
of the program. PL/I has a set of ON-conditions corresponding
to these exceptional conditions. When the condition occurs,
it is said to be raised. Using an ON-statement , the programmer
can specify a response to the condition in the form of a
statement (or begin-block) to be executed. That response is
known as the ON-unit. If no ON-unit has been specified, a
standard ON-unit, known as the standard system action, is
executed. Depending on the nature of the ON-condition, it may
be possible for the program to continue where it left off after
the ON-unit is executed. The different kinds of ON-conditions
are listed in Table 7.
An example illustrating the use of ON-conditions and
ON-units is:
P : PROCEDURE ;
ON ENDFILE(SYSIN)
GO TO PROCESS;
DO WHILE ( 'I'B) ;
READ FILE (SYS IN) INTO (LINE IMAGE);
PROCESS:
END;
END P;
102
c
u
3
-H O
o
2
t/3 (/) U) WW
(U (U Q) d) <]) O
>i >, >, >, >, C
w
en
w
U)
q;
o
0)
Q)
0
Q)
>i
c
>i
>,
c
>i
+J
C
r.
0)
0)
e
rH
(U Xll
1— 1
•rH
XI
W
nj
03
c
O
u
04
o
c
o
c
to
o o
c c
o
c
(0
o <u
o
c
n
o <u
c >,
o
c
e
0)
0)
c
4J
0
U)
•H
>
4J
w c
•H
0
-a
13 -H
c
V^ +J
o
ITJ O
u
-a <
1
c
2
nj
o
-p
w
El
0)
JQ
(0
c
■H
tn
4J
c
i o
o a:
o w
•H 4J
0
(0 iH
0)
.5^
c
C 0)
OH
■H O
ig 14-1
O D
lO
■p
C
O
O
0)
M 0)
(0
03 O
1— 1
3
fl
10
c
u
^ w
a>
0)
U
O
a
c
•H
tn
-P
c
0) a:
i o
e a
o a
u u
I
o
E
O
0)
4J
V^
>
c
o
u
0)
a
o
a
c
C (U
•H a
(1) >i
XI -P
3 (1)
m o
> 03
o
a
w
ITJ
C
•H
CO
c
(U
o
o
4J
3
a
c
c
o
ro
■P
TJ
c
4J
)-l
ITJ
CO
r-i r-i
•H -H
4J
c
■H
)-l
a
c
o
0)
CP
ITJ
a
T3 73
C C
01 <u
I
c
o
■H
4-)
c
e (1)
Q) C
a>4-i
•H XJ
(1)
E
-rH
4J
I
c
o
■H
-P
3
U
0)
X
0)
3
O
0)
c
n3
V)
u
o
u
u
c
o
•H
o
IT3
c
o
•H
4J
3
O
(U
X
0)
E
it3
M
CTi
O
U
a
c
0
o
(TJ
c
en
•H
tn
-P
c
a)
o o
c V
I
(0 -o
U 0)
(U -P
a-u
o -H
E
1-1
0)
a
u
-H
4J
0)
E
x:
>-l X
(TJ (T3
E
-a
<u en
X c
•H -H
M-l TJ
OJ
U
X
0)
O -t
3
0)
0)
3
o
ITJ
C
-p
c
0)
o
u
0)
>;
0) TI
rH M
XI O
(0 U
-P (U
a >-i
0)
o ty
u c
H3 -H
c u
3 3
T)
M
O T3
0) M -P
4J
3
ITJ -P
O
3
O
C I
3 -P
Wi 4J >
O 3
a o a
3 c c
no 0) H
o
c
en
•H
T)
C
(TJ
^J
c
Q)
E E
E (1)
O -P
U H
TI
Q)
U
0)
■P
C
3
O
o
c
3
a
c
■H
0) (TJ
(1)
0) Vj
E -P
(tj in
c
TJ
-O 0)
0) -P
N O
•H 0)
C ^1
O^-H
O T)
O I
0) (T3
U -P
C ITJ
3 T)
Pi
o
Oi
(TJ
c
t7^
C
o
u
I
(0
(1)
a
o
QJ
E
x:
E -P E
H -H 3
M X ^
(Tj (TJ (T3
E >
4J
ITJ
O
C 0)
>*-i TJ XJ
0) IT3
0) -P
O
X
0
3
in
k4
(1) U)
o
a:
u
ITJ
c
•H
W
4J
c
OJ
o
o
x:
ty u
c o
Q) O
4J
0) c
u
u
3
O 73
C -P
•H TJ 3
3
o
I
3 -P
O 3
a
c
c
o
•H
■p
■H
•o
c
o
u
<
<
Z H
O >•-'
M -H
Eh ^J
M C
a 0)
2 T3
O H
u —
z
o
V)
a:
>
2
O
o
Qi 0)
•H H
<4H >«H
U U
<
Q
2
a.
Q
2
U Cd
o
a:
u
X Q
to Q
t-H U
2 X
103
0)
U
in:
0)
Cm
a:
>
o
o
o
u
c
u
(U Q>
o
e <
o
z
w
m
w
U)
w
0
0)
0
(U
0
<u
0)
0)
0
c
>1
c
>l
c
>1
>1
>1
c
0)
C
-H
-P
c
o
u
-p
c
t^.
Q)
<D
e
<-{
0) JDI
rH
■H
XJ
W
(0
tn
c
0
w
04
e
<D
+J
U)
>1
en c
o
V4 -P
(U U
TD <
C
(0
-P
cn
o
a
a
w
m
c
en
-H
CO
4J
C
(U
o
o
(0
o <u
C >i
0) a
c o
I
c
o
-H
-p
(0
-p
c
e
(0
c
en
c
m
(1)
>1
CI. g
e o
■H O
c
o
-p
o
rO
o
c
en
0) o
o o
C C
cn en
•H -rH
-p -P
c c
0) Q)
O O
o
c
o
a:
u
(0
c
cn
■H
cn
c
o
u
(0
>1
■p
c
0)
o
u
cn
0)
(X
o
«
Pi
u
03
c
tn
c
0)
o
u
0)
Xi
<a
hi
c
o
-p
•H
c
O
U
>i
■P
-H
U
(C
Q.-P
03 C
U QJ
E
cn c
QJ
-a
cn
cn
QJ
■rH
3
0)
cn
(0
O
cn
U
X
03
M-l
a) o
■H -P
03 (U
> cn
s-i
03
-p
T3
<])
X
■rH
o
w
M
cn
dJ
-P
cn
3
m
x:
X
OJ
(U
cn
03
>-i
O
-P
cn
03
03
>
03
Cn
C
cn -rH
u -p
c cn
0)
S-i 03
(U
m 4-1
(1) o
^
c
o
cn -p
CQ u
D O
cn a,
M-l -P
o c
(1)
■P 4J
C CO
Q) -H
X
0)
cn c
h o
03 C
e
3
W S
2
o a:
Eh Eh
cn cn
-p
cn
c
cn
T) -P
(U C
(U 0)
O E
X C
CD tr
■H
cn cn
c cn
-H 03
>-(
cn o
M-l +J
O Q)
cn
s: U
-P 03
Cn-P
C
(U MH
^ O
w
to
H
cn
o
2
M
cn
cn
c
03
-p
3
O
-P
a
•H
U
cn
X)
3
cn
w
ID
2
Eh
&4
cn
c
-rH
3
'O
u
o
>-l
S-I
0)
c
o
cn
cn
•H
E
cn
c
03
u
4J
c
o
■H
4-1
03
Sh
<U
a
o
-p
3
■P
3
O
I
4-1
3
c
•H
cu
4-1
cj cn
cn 2
00 Eh
Q)
C
(D
o
-p
o
c
c
03
U
0)
iH
•H
m
w
M
Q
W
2
M -^
W .H
Q -H
2 4-1
ID ^
104
I
o
o
■H E
-P 3
OJ E
E -H
ji: c
■P -rH
■H E
!-l
03
0)
3
1— t
05
>
X!
03
O 03
-P
C
(U
cn cn
QJ
C U
o a.
H 0)
>-( +j i-i
o
w
Q
2
D
O
Jh
<U
N
>i
XJ
c
o
c
03
-p ^ QJ cn
03 -P
O
m QJ 4J
rH OS
T3
QJ
-P
e
QJ
4J
-p
03
W
Q
o
o a
M W
cn >
(X o
>
2
O
U &H
cn
c
o
Xi
c
o
o
n
QJ
rH
ja
m
c
0)
o
Q Cn
w a;
w
IS!
M
cn
a
2
X w a
> H
o cn
w
Q
Q
o
w
isi
a
2
o
2
2 U
w M cn
N a cQ
M Eh 3
cn cn cn
M 03
Q -H
O -P
£X -H
U C
CS] M
0)
C
o
•H
■o
c
o
o
TJ
OJ
XI
m
0}
•H
T3
>1
01
•rl
-p
■rl
C
In this case, the ON ENDFILE statement specifies that when
end-of-file is encountered on the file SYSIN, the statement
GO TO PROCESS; is to be executed. Thus the file will be read
up to its end, and afterwards the statements at PROCESS will
be executed.
The ON-Statement , REVERT-Statement , and SIGNAL-Statement
The ON-statement specifies a list of ON-conditions
together with an ON-unit. The association of the ON-unit with
the ON-conditions is not made until the ON-statement is
actually executed. Moreover, execution of a subsequent ON-
statement can supersede the effect of an earlier one. For
instance, after execution of the two ON-statements
ON OVERFLOW, FIXEDOVERFLOW
GO TO TOOB IC-
ON FIXEDOVERFLOW
GO TO FIXEDBIG;
the ON-unit associated with the OVERFLOW condition is
GO TO TOOBIG;
while the ON-unit associated with the FIXEDOVERFLOW condition is
GO TO FIXEDBIG;
ON-statements have block scope, in the sense that they are
effective only until the block containing them is terminated.
When execution of a block is completed, the association between
ON-conditions and ON-units reverts to what it was in the previ-
ously-executing block. Thus a procedure can activ^* • t collection
105
of ON-units appropriate to its circumstances without affecting
the ON-units set up by its caller.
The ON-unit itself can be either a BEGIN -block (delimited
by BEGIN and END) or a single unconditional statement. In
particular, an ON-unit cannot be either a DO-group or an IF-
statement. Actual execution of the ON-unit is carried out as
though the ON-unit were a procedure. In particular, ON-units
carry environments with them, and so any names occurring in an
ON-unit have the meaning applicable at the point of execution
of the corresponding ON-statement .
An ON-statement can specify the standard system action as
an ON-unit, using the keyword SYSTEM. Thus
ON SUBSCRIPTRANGE
SYSTEM;
specifies that the standard system action is to be taken if the
SUBSCRIPTRANGE-condition is raised. This facility can be used
to nullify the effect of previously executed ON-statements . It
is also possible to specify that a traceback , or other debugging
information, is to be printed in the event that a condition is
raised. That effect is gotten by using the SNAP keyword, as in
ON SIZE SNAP SYSTEM;
If the SIZE-condition is raised, the standard system action will
be taken, but in addition debugging information will be printed.
The actual choice of debugging information is implementation-
defined. The statement
ON SIZE SNAP;
106
would produce a different effect: if the SIZE-condition is
raised, the null-statement, which does nothing, will be executed,
In fact, the null-statement is not a valid ON-unit for the
SIZE-condition because it does not terminate in a GOTO-
statement. The question of validity of such ON-units is
discussed below.
The REVERT-statement can be used to cancel the effect of
an ON-statement , or several of them, without knowing what ON-
condition was in effect previously. The REVERT-statement speci-
fies a list of ON-conditions . Execution of the REVERT-statement
causes the ON-unit for each of these conditions to revert to
what it was in the previously-executing block. Thus, in the
sequence:
ON ENDFILE(SYSIN)
CALL ENDER;
BEGIN;
ON ENDFILE(SYSIN)
GO TO ALT_END;
REVERT ENDFILE(SYSIN) ;
END;
the REVERT-statement causes the on-condition
CALL ENDER;
to again be associated with the ENDFILE-condition for the file
SYSIN.
The SIGNAL-statement is used to raise a specified ON-
condition. For example, the statement
SIGNAL ZERODIVIDE;
107
causes the ZERODIVIDE-condition to be raised and the appropri-
ate ON-unit (possibly the standard system action) to be invoked.
This statement is particularly useful in debugging program logic
for handling ON-conditions . It is also the only way to raise
a programmer-defined condition (discussed below) .
Enablement and Disablement
A number of the ON-conditions require time-consuming code
(on most machines, at least) in order to check whether or not
they have occurred. The time needed to check whether a sub-
script is out of range, for instance, well may dominate the
time needed for the retrieval of a subscripted variable. There-
fore PL/I allows the programmer to either enable (turn on) or
disable (turn off) the check. Enablement and disablement are
provided only for certain ON-conditions. They are specified by
means of a condition prefix, which consists of either an ON-
condition name or the negation of an ON-condition name, in
parentheses and followed by a colon. The condition prefix can
be applied either to a single statement or to a block. For
example, in the sequence:
(OVERFLOW, NOS I ZE) :
P: PROCEDURE;
(NOOVERFLOW) : Q = A + BTR ( I ) ;
END;
the SIZE-condition is disabled throughout the procedure P, while
the OVERFLOW-condition is enabled throughout P except for the
108
single statement where NOOVERFLOW is indicated. For that
statement, the condition is disabled, and no test will be made
for it. If a condition is raised in a statement where it
has been disabled, that is considered to be a programmer error,
and the implementation is not to be held responsible for its
consequences .
Enablement and disablement are static properties of a
program. In other words, it is possible to tell whether a
particular ON-condition is either enabled or disabled for a
particular statement just by looking at the program, without
considering what its sequence of execution is. In this respect,
enablement and disablement differ from the ON-statements ,
whose execution depends on program flow. Enablement and dis-
ablement affect whether a condition is or is not to be tested
for, while ON-statements determine what action is to be taken
if the condition is raised. The statement
ON UNDERFLOW;
does not disable the UNDERFLOW-condition; it merely states that
if that condition is detected, the null-statement is to be
executed.
Builtin Functions for ON-Conditions
During the execution of an ON-unit, a number of builtin
functions are available in order to determine the circumstances
that caused the corresponding ON-condition to be raised. Some
of these apply to all ON-conditions and are discussed here;
109
other are specific to particular ON-conditions and are
discussed in connection with those conditions. In general,
these builtin functions do not have meaningful values except
in the context of an ON-unit. They are all functions of no
arguments .
The ONCODE builtin function has as its value an imple-
mentation-defined integer used to indicate why the active ON-
condition was raised. A particular condition may have more
than one code value associated with it. One common convention
is that the value of ONCODE is zero if the ON-condition was
raised by a SIGNAL-statement . The ONLOC builtin function
returns as its value the name of the innermost entry point
active when the condition was raised. For input-output-related
ON-conditions, the ONFILE builtin function has as its value
the name of the file that was being operated upon when the con-
dition was raised. The values of both ONLOC and ONFILE are
in the form of character strings.
Categorization of the ON-Conditions
The various ON-conditions listed in Table 7 can be broken
down into three groups. The first group consists of the compu-
tational ON-conditions. Most conditions in this group are
raised in response to a particular kind of error. The computa-
tional ON-conditions are the only ones that can be enabled
and disabled. They are:
110
CONVERSION
FIXE DOVE RFLOW
OVERFLOW
SIZE
STRINGRANGE
STRINGSIZE
SUBSCRIPTRANGE
UNDERFLOW
ZERODIVIDE
The occurrence of one of these conditions usually means that a
bad result has been generated, and so the active computation
cannot be continued. For this reason, the ON-units associated
with most of these conditions must not terminate normally,
i.e., must cause a transfer of control out of the ON-unit by
means of a GOTO- statement or similar construction. Normal
termination would mean that the active computation would be
resumed, and the nature of the condition is such that the
computation cannot be resumed. For instance, if a subscript
is out of range on an array reference, there is no way to
obtain an appropriate value for the reference.
Three of the computational ON-conditions are treated
somewhat differently. The CONVERSION-condition is raised when
data is being converted from character to some other type.
When this condition is raised, the programmer can modify the
character string to be converted. If a normal return takes
place from the ON-unit, i.e., the ON-unit completes without
a transfer of control, the conversion is reattempted with the
modified input string.
Two builtin functions are available for the modification:
ONSOURCE and ONCHAR . ONSOURCE has as its value the character
string to be converted, while ONCMAR has as its value the left-
Ill
most character in that string for which no valid continuation
exists. By examining ONSOURCE and ONCHAR, the programmer may
be able to determine the difficulty and what to do about it.
Moreover, ONSOURCE and ONCHAR can be used on the left side of
an assignment (within an ON-unit for the CONVERSION-condition) ,
and so the string to be converted can be modified by assign-
ments to either ONSOURCE or ONCHAR (which can also be used as
pseudovariables) . For example, if a character string is being
converted to a bit string the following ON-unit might be
appropriate:
ON CONVERSION BEGIN;
DECLARE ONCHAR BUILTIN;
IF 0NCHAR='fe5' THEN
ONCHAR = • 0 ' ;
ELSE ONCHAR = ' 1 ' ;
END;
If the string to be converted does not consist entirely of ones
and zeros, each blank in that string will be replaced by a zero,
and each other deviant character will be replaced by a one.
The UNDERFLOW-condition also receives slightly different
treatment. Normal return from the UNDERFLOW-condition is
permitted, and the value of the computation that underf lowed is
taken to be zero. The STRINGSIZE-condition arises when a string
is shortened as a result of a conversion or assignment. Upon
normal return from the ON-unit, the string is truncated on the
right to the required length. Since the standard system action
in this case is to do nothing, this condition is often ignored.
However, it can be used in either of two ways. If it is disabled,
then the compiler need not produce code to check for string
112
overflow. Moreover, if a nonstandard ON-unit is provided,
then the programmer can take some action. However, there is
no way that the programmer can modify the result produced
either for the UNDERFLOW-condition or for the STRINGSIZE-
condition .
The second group of ON-conditions is the input-output
conditions. Each of these conditions is associated with a
particular file, specified along with the condition name.
The input-output conditions are:
ENDPAGE
ENDFILE
KEY
NAME
RECORD
TRANSMIT
UNDEFINEDFILE
Some of these conditions are discussed further in connection
with input-output.
The remaining conditions are more varied. These are:
AREA
CONDITION
ERROR
FINISH
STORAGE
The AREA-condition is raised when an allocation is attempted in
an area, and there is insufficient space for the allocation.
If the associated ON-unit returns normally, the area-reference
in the ALLOCATE-statement is reevaluatd, and the allocation is
reattempted. Therefore an appropriate response to the AREA-
condition is to assign a new area value to the area variable
referenced in the ALLOCATE-statement.
113
The programmer can define ON-conditions using the keyword
CONDITION and an identifier, known as the condition-name , ON-
units can be provided for programmer-named conditions, but they
can only be raised by a SIGNAL-statement . For instance, a
programmer might write:
ON CONDITION (TABLE_OVERFLOW)
CALL OVERFLOW_RECOVERY ;
and then, in some other part of the program, write:
IF T > TABSIZE THEN
SIGNAL TABLE_OVERFLOW;
The ERROR-condition is raised under a variety of circum-
stances, some of which can be implementation-defined. The
standard system action in response to a number of other ON-
conditions is to comment (i.e., display diagnostic information)
and then to raise the ERROR-condition. (It is quite acceptable
to have an ON-unit raise an ON-condition itself.)
The FINISH-condition is raised when the program completes.
It differs from all other conditions in that it is raised as
a normal aspect of program execution. The STORAGE-condition is
raised when the program runs out of storage. Since programs
consume storage in many different ways, the exact circumstances
under which it is raised are implementation-defined. Recovery
from this condition may or may not be possible.
114
OTHER STATEMENTS AFFECTING FLOW OF CONTROL
Conditional Statements
The conditional statement is used in order to test a
condition and take some action depending on the result. A
conditional statement starts with an IF-statement , specifying
the test, and may include an ELSE-part that specifies what
action to take if the test fails. For instance, the sequence
IF Q <= QMAX THEN
INDEX = INDEX+1;
ELSE
GO TO PART_7;
causes the assignment
INDEX = INDEX+1;
to be executed if the condition Q <= QMAX is true, and the
statement
GO TO PART_7
to be executed otherwise. The statement following either THEN
or ELSE can itself be a conditional statement, so that nests
of conditional statements can be built up. Moreover, either
THEN or ELSE can be followed by a DO-group (discussed below) ,
so that several statements can be executed after the test
rather than just one. An example of a more complicated
conditional statement is:
115
IF A(I)=0 THEN DO;
SIZE1=SIZE1+INCR;
SIZE2=SIZE2-INCR;
IF SIZE2<SIZE1 THEN
CALL ADJUST;
END;
END IF A (I) >0 THEN
SIZE2=SIZE2+INCR;
ELSE
SIZE1=SIZE1-INCR;
It is not necessary that each IF-statement have a corresponding
ELSE- statement . In complicated conditional statements, each
ELSE is paired with the nearest preceding unpaired IF, working
from front to back.
The test in an IF-statement actually takes the form of
an expression, which is evaluated and converted to a bit
string. Since the comparison operators all produce one-bit
results, and since the logical operators also produce one-bit
results when their operands are one-bit values, the conversion
is usually unnecessary. If the bit string obtained by
evaluating the test expression has at least one one-bit in it,
the test succeeds, and otherwise it fails. The test expression
must be scalar-valued, although if it is not scalar-valued the
SOME and EVERY builtin functions can be used to reduce it to
a scalar value .
The DO- Statement
The DO-statement has three main variants: the simple DO,
the DO-WHILE, and the specified DO. The simple DO is used in
order to convert, syntactically, a sequence of statements into
a single statement. A simple DO-group has the form:
116
DC-
Statement- sequence
END;
The statements in the sequence are executed just once. Transfers
of control into and out of the sequence are permitted. The main
use of the simple DO-group is as part of a conditional statement,
The DO-WHILE variant has the form:
DO WHILE (expression) ;
A DO WHILE-group consists of a DO WHILE-statement followed by
a statement sequence and a matching END-statement . The state-
ments in the group are executed repeatedly, and the expression
in the DO WHILE-statement is tested before each execution. If
the test fails, control is transferred to the statement follow-
ing the group. If the expression is initially false, the group
is not executed at all. An example of a DO WHILE-group is:
DO WHILE (CVAL>0) ;
DVAL=DVAL+G(CVAL) ;
CVAL=CVAL-DVAL ;
END;
The specified DO itself has a number of variants. As with
the other two forms, a DO-group consists of a specified DO-
statement followed by a statement sequence followed by an END-
statement. The most common variant is illustrated by:
DO M = 0 TO 100 BY 2;
In this case the statements in the group are executed repeatedly.
Before the first execution, M is assigned the value 0. M is
then increased by 2 on each execution of the group, and has
117
the value 100 on the last execution of the group. Upon comple-
tion of the entire group, the value of M is 102. However,
transfer out of the group is permitted, and if that happens,
M retains the value assigned to it on the most recent iteration.
The TO-clause and the BY-clause can be written in either
order, and either of them can be omitted. If the TO-clause is
omitted, the group is iterated indefinitely, i.e., until a
transfer of control out of the group takes place. If the BY-
clause is omitted, a value of 1 is assumed for it. On each
iteration, the control variable (the variable following the
keyword DO) is incremented by the value given in the BY-clause.
If the BY-clause has a negative value, then the control vari-
able is decremented rather than incremented. The loop terminates
when the value of the control variable is greater than the
value of the TO-clause (for a positive BY-value) or less than
the value of the TO-clause (for a negative BY-value) . If the
termination test is satisfied by the initial value of the
control variable, the group is executed zero times. If neither
the TO-clause nor the BY-clause appears, the group is executed
for a single value of the control variable.
The TO-clause and the BY-clause are both evaluated prior
to execution of the statements within the DO-group. Thus any
changes to values of variables that appear within the TO-clause
or the BY-clause have no effect once the iteration has started.
The control variable need not have arithmetic type; a string or
pictured type is also acceptable. A WHILE-clause can also be
specified, e.g . ,
118
DO JV = X BY Y WHILE (PROP (JV)< PROP (JV+1) ) ;
Another variant is illustrated by:
DO P = LIST_HEAD REPEAT (P->NEXT) WHILE (P"'=NULL) ;
or
DO STRING='' REPEAT (STRING I I CHARS (I) ) WHILE (LENGTH (STRING) < LMAX) ;
The control variable is assigned the given initial value on
the first iteration. On subsequent iterations, the value of
the REPEAT-clause is recalculated and assigned to the control
variable. The WHILE-clause can be omitted, although usually
it is desirable to include it.
The specified DO can consist of a sequence of specifications
rather than a single one. For instance, the statement
DO M = 3,7,M+2 BY 3 TO 15,0;
executes the group of statements that follows for the sequence
of values 3, 7, 9, 12, 15, 0. Each specification in the group
can have the general forms described above.
The GOTO-statement
The GOTO-statement causes control to be transferred to
the label specified in the statement. The statement actually
specifies a label-valued expression, and although that expres-
sion normally is a constant, i.e., a statement-name, it need
not be. For instance, it could be a subscripted reference to
an array of statement-names, so that the appropriate destina-
tion is selected by the value of an index.
119
The destination of a GOTO-statement need not be in the
same block as the statement itself. If the destination is
in a different block, then the effect of the statement is to
terminate execution of the current block and all blocks
between the statement and its destination. In other words,
at the moment when the GOTO-statement is executed, there will
be a hierarchy of active blocks, with the current block last
in the hierarchy. The label value obtained from the GOTO-
statement must designate, as its environment component, some
block in the hierarchy. Then all blocks between the designated
block and the current block, as well as the current block
itself, are terminated. The designated block then becomes
the current block, and control is transferred to the statement
named by the label value.
A GOTO-statement whose destination is not in the same
block as the statement itself is called a nonlocal goto.
A nonlocal goto is expensive to execute relative to a local one.
Therefore the programmer is allowed to declare the LOCAL-
attribute for a label variable. The LOCAL-attribute consti-
tutes a claim by the programmer that any GOTO-statement using
the value of that label variable will be a local goto. Thus
the compiler need not examine the environment associated with
the label, and can generate instructions to execute the
requested transfer of control directly.
120
The STOP-Statement and the Null-Statement
The STOP-Statement has the form
STOP ;
and is used to stop execution of the program. It has the effect
of terminating the execution of all currently active blocks.
The null-statement has no text at all; it is written as
just a semicolon. Its main uses are to place a statement-name,
to fill out a branch of a conditional statement where no action
is to be taken, and to specify that no action is to be taken in
response to a specified ON-condition .
121
FILES AND RECORD INPUT-OUTPUT
File Attributes
The attributes of a file determine the kinds of operations
that can meaningfully be applied to that file. Moreover, they
dictate to some extent the characteristics of the dataset
associated with the file. The final determination of file
attributes takes place when the file is opened, i.e., associ-
ated with a dataset. If a file is opened and closed several
times, it can have different attributes at different openings.
The file attributes INPUT, OUTPUT, and UPDATE determine
the direction of information flow in an obvious way. An opened
file must have either the RECORD-attribute or the STREAM-
attribute. A record file is associated with a dataset con-
sisting of a sequence of reaords , which are read or written
as single units. A record may or may not have a key associ-
ated with it. If it does, then the file has the KEYED-
attribute. If the records are sequenced, i.e., the notion of
"next record" is meaningful, then the file has the SEQUENTIAL-
attribute; otherwise it has the DIRECT-attribute . A direct
file is necessarily keyed, since without a key there is no
way to designate a record within the file, while a sequential
file may or may not be keyed.
A stream file is associated with a dataset consisting of
a sequence of characters. Within the sequence of characters,
linemarks , pagemarks , and carriage-returns can appear. A
linemark marks the break between the characters on two succes-
122
sive lines (either input or output) . A pagemark is meaningful
only for a dataset that is to be printed, and marks the start
of a new page. A carriage-return is meaningful only for a
dataset that is to be printed, and indicates that the follow-
ing line is to be overprinted, i.e., printed without advancing
the carriage. A file associated with a dataset to be printed
has the PRINT-attribute , and consequently the OUTPUT-attribute
also.
A file may also have an ENVIRONMENT-attribute associated
with it. The ENVIRONMENT-attribute contains implementation-
defined information describing the associated dataset. Typical
items found in the ENVIRONMENT-attribute are record length,
blocking factors, record formats, character set selection, and
tape densities.
File Opening and Attribute Determination
A file can be opened either explicitly, through execution
of an OPEN-statement, or implicitly, through execution of some
other input-output statement referencing an unopened file. When
the file is opened, any attributes given in the file declara-
tion are combined with those given in the OPEN-statement. The
resulting partial set of attributes is then checked for consis-
tency. If any inconsistency is found, the UNDEFINEDFILE-
condition is raised for the file. Then defaulting rules are
applied to generate a complete set of file attributes. The
possible sets of file attributes are given in Table 8. For
123
TABLE 8. Complete Sets of File Attributes
STREAM INPUT FILE
STREAM OUTPUT FILE
STREAM OUTPUT PRINT FILE
STREAM INPUT SEQUENTIAL FILE
RECORD INPUT SEQUENTIAL KEYED FILE
RECORD INPUT DIRECT KEYED FILE
RECORD OUTPUT SEQUENTIAL FILE
RECORD OUTPUT SEQUENTIAL KEYED FILE
RECORD OUTPUT DIRECT KEYED FILE
RECORD UPDATE SEQUENTIAL FILE
RECORD UPDATE SEQUENTIAL KEYED FILE
RECORD UPDATE DIRECT KEYED FILE
Note: The ENVIRONMENT-attribute may be added
to any of these combinations.
124
example, in the sequence
DECLARE CHANGES INPUT FILE;
OPEN FILE (CHANGES) KEYED;
the initial set of attributes used for opening the file CHANGES
is INPUT KEYED. Since these attributes are consistent with
each other, the opening can proceed. The RECORD and FILE
attributes are implied by the KEYED attribute, and so these
are added to obtain the set RECORD INPUT KEYED FILE. Although
both DIRECT and SEQUENTIAL are consistent with this set, the
default choice is SEQUENTIAL, and so SEQUENTIAL is added to
obtain the complete set
RECORD INPUT SEQUENTIAL KEYED FILE
The FILE-option in an OPEN-statement contains a file-valued
expression, which is either a constant or something that
evaluates to a file constant. It is assumed that each dataset
has a name (typically, known to the surrounding operating
system) and so the file opening has to specify the name of the
dataset to be linked to the file. The name can be specified
by a TITLE-option; if it is not, the name of the file constant
obtained by evaluating the FILE-option is used. Thus the
statements
DECLARE CHANGENAME CHARACTER (4 ) ;
CHANGENAME = '04 37' ;
OPEN FILE (CHANGES) TITLE (CHANGENAME) STREAM INPUT;
OPEN FILE(OLDSET) RECORD INPUT DIRECT;
cause the file CHANGES to be associated with the dataset C4 37,
and cause the file OLDSET to be associated with the dataset OLDSET.
125
For stream files, other information can be given in the
OPEN-statement . For instance,
OPEN FILE(LISTING) PRINT LINESIZE (110 ) TAB ( 10 , 40 , 70 )
PAGESIZE(50) ;
causes the file LISTING to be opened with the understanding
that a new line will be started after at most 110 characters,
a new page will be started after at most 50 lines, and tabstops
(discussed under "Edit-Directed Input-Output" below) will be
placed at column positions 10, 40, and 70. Were any of these
values to be omitted, implementation-defined values would be
assumed.
If an input-output statement is executed on a closed file,
then the file is implicitly opened. For instance, if the
statement
PUT FILE (ANS) (M,N) ;
is executed and the file ANS is not open, the file will be
opened with the implicit attributes STREAM and OUTPUT.
Similarly, if the statement
DELETE FILE(INV) KEY (PART_NAME) ;
is executed and the file INV is not open, it will be opened with
the implicit attributes RECORD and UPDATE. The implicit attri-
butes are treated as though they appeared on an OPEN-statement,
so any attributes given in the declaration of the file are
combined with those derived from the implicit opening.
126
File Closing
Just as the OPEN-statement creates the connection between
a file and a dataset, the CLOSE-statement breaks the connection.
All files are automatically closed at program termination. The
programmer can also specify actions such as dataset disposition
by means of an ENVIRONMENT-attribute attached to the CLOSE-
statement, e.g.,
CLOSE FILE(BIBLIO) ENVIRONMENT (REWIND) ;
A file can be close! and later reopened with different attributes
An attempt to open a file that is already open, or to close a
file that is closed, has no effect.
Operations on Record Files
There are five statements applicable to record files:
READ
WRITE
LOCATE
REWRITE
DELETE
Each of these, in turn, has a number of clauses that can be
applied to it. The attributes of a file determine which state-
ments and clauses can meaningfully be applied to that file.
For instance, READ cannot be applied to an output file, DELETE
can only be applied to an update file, and any clause that
references a key can only be applied to a keyed file. The
meanings of the statements are summarized in Table 9, and the
meanings of the clauses are summarized in Table 10. Table 11
127
Table 9. Record Input-Output Statements
READ
REWRITE
WRITE
LOCATE
DELETE
read a record from a dataset
(input and update files only)
replace a record on a dataset
(update files only)
add a record to a dataset
(output and update files only)
obtain buffer space for a record
(output files only)
delete a record from a dataset
(update files only)
Table 10. Clauses on Record Input Output Statements
FILE
INTO
FROM
KEY
KEYFROM
KEYTO
IGNORE
SET
specifies the file accessed by this statement
specifies the generation to receive a record
being read
specifies the generation containing a record
to be written
specifies a record to be read, replaced,
or deleted
specifies the source for a key to be attached
to a record to be written
specifies where to put the key associated
with a record being read
specifies the number of records to be skipped
by execution of a read statement
specifies where to put a pointer to a newly
created generation
128
<
Q
3
Q
X
X
X
k J
•^
X
-
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
in
X
X
X
X
X
X
X
X
X
X
Eh
D
Eh
D
O
Q
X
X
C/3
X
X
w
X
X
Eh
D
cu
2
M
Q
X
X
X
X
CO
it:
X
X
X
X
X
X
X
X
X
w
X
X
X
c
■P
ro
4-1
W
«w
O
E
0
z
M
u
M
Q
<
>>
u
u:
z
M
14-1
U
M
Q
<
>H
u
:<:
o
£h
Z
M
M
Q
a
Eh
u
w
u-i
u
M
Cm
a
<
>H
u
:<:
a
Eh
u
w
u
M
u.
Q
<
u
t<
a
Eh
U
W
U-l
►J
M
Ci4
Q
<
u
o:
o
2
U
M
u
J
M
Ci4
Q
<
u
u
a:
o
o
z
M
U
M
Q
M
O
Eh
>H
U
o
z
o
M
X-l
U
M
Cl,
Q
<:
-
o
a:
<4H
u
►4
M
Cm
CO
Eh
M
q:
2
i
PS
Cm
U
a:
Cm
CO
►4
M
Cm
M
2
a
Eh
to
Cl)
M
Cm
>
Eh
<
8
Cm
>^
CJ
a
Eh
CO
en
CO
►J
M
Cm
>
CO
Eh
<
M-l
CO
M
Cm
CO
Eh
i-1
§
Cm
Im
CO
M
Cm
CO
Eh
t— 1
CO
">
§
Cm
U-l
CO
.4
M
Cm
U-l
CO
M
Cm
CO
:«:
U-l
CO
M
Cm
CO
Eh
CO
(U
r-i
JQ +>
<a c
(1)
•H 3
eH
U
H o
Xi
Q)
(0 o
nj
4-"
>
•H Q)
C
a
M r-{
-H
>
>i-H
(T3 -H
O
0)
0) M
> IW
aj^
M (0
> HH a^ j«i c
c
o
■M
4-1
(0
)-i
0)
a
o
0)
m oj
a
>1
E
0)
a d)
<TJ
•0 1-
c
•o n cr
c o o»
(0
•H
4J
C *J
0) u
3 Q>
cr u
0) -H
0) T3
>i >»
0) 0)
to o
tn u: »c
129
shows how the different statement forms relate to the
different combinations of file attributes.
The chief characteristic of record input-output is
that it involves a direct transfer of information between
the dataset and addressable memory, without any formatting
or editing. It is therefore the programmer's responsibility
to be sure that the format of information on the dataset
agrees with the format in memory, as defined by the imple-
mentation. If the dataset is itself created by a PL/I pro-
gram, this is not too difficult.
The READ- statement causes a single record to be read
from a specified file. It can also be used to skip records
on a sequential file. Its simplest form, applied to a
sequential file, is illustrated by
READ FILE (CUST) INTO (CUST_INFO) ;
A single record is transferred from the dataset associated
with CUST into the variable CUST_INFO. If the size of the
record disagrees with the size of the variable, the RECORD-
condition is raised for the file. If there are no more records
left in the dataset, the ENDFILE-condition is raised for the
file.
The destination of a newly read record can be specified
either by an INTO-clause or by a SET-clause. If the INTO-
clause is given, then the record is read into a buffer, and
the pointer specified in the SET-clause is set to the location
of the record within the buffer. Although the SET-clause is
130
less convenient to use than the INTO-clause, it has two
advantages. First, the SET-clause can be used to read
records whose length is specified within the record itself.
The INTO-clause cannot be used for such records because the
size of the necessary variable is not known. Second, operat-
ing on the record in the buffer avoids the need to copy
the record from the buffer to the variable.
The KEY-clause is used to specify the position within
the dataset of the record to be read. For instance, the
statement
READ FILE (EMPL) INTO (EMPLOYEE_RECORD) KEY (EMP_NUMBER) ;
causes the record whose key is the character-string form of
EMP-NUMBER to be read into the variable EMPLOYEE_RECORD.
(If the key value is not a character string, it is converted
to one.) If the key designates a nonexistent record, the KEY-
condition is raised for the file.
The KEYTO-clause, in contrast, does not influence the
dataset position at all. Instead the key in the record being
read is assigned to the variable specified in the KEYTO-clause.
This facility is necessary since the key may not be part of
the record itself. For instance, the statement
READ FILE (EMPL) INTO (EMPLOYEE_RECORD) KEYTO (EMP_NUMBER) ;
causes the next record from EMPL (which must be keyed sequential)
to be read into EMPLOYEE_RECORD, and the key associated with
that record to be assigned to EMP NUMBER.
131
The IGNORE-clause can be used in order to skip records.
For instance, the statement
READ FILE(EMPL) IGN0RE(2);
causes two records on the dataset associated with EMPL to be
skipped. (EMPL must necessarily be sequential.) The IGNORE-
clause and the KEY-clause can be used together, and if the
count in the IGNORE-clause is zero the effect is to position
the dataset at the record designated by the KEY-clause. The
IGNORE-clause and the KEYTO-clause can be used together to
read a key without reading the associated record.
The WRITE-statement can be used only with output or direct
update files. It causes the variable named in the FROM-clause
to be written onto the dataset. If the dataset is sequential,
the new record is written at its end; otherwise the position
of the new record is arbitrary. The KEYFROM-clause, which
specifies the key of the new record, must be used if the file
is keyed.
The LOCATE- statement can be used only with output files.
It is analogous to the READ- statement with the SET-option.
Execution of a LOCATE-statement causes space for the variable
named in the statement to be allocated in an output buffer,
and then causes the pointer named in the SET-clause to be set
to the location of this space. If the variable is declared
with a pointer, then the SET-clause can be omitted and the
pointer obtained from the declaration will be used. For
instance, given the statements
132
DECLARE RECSPACE CHARACTER ( 20 0) BASED (CHPTR) ;
DECLARE CHPTR POINTER;
the statements
LOCATE RECSPACE FILE (NEWSET) ;
and
LOCATE RECSPACE FILE (NEWSET) SET (CHPTR) ;
are equivalent. After execution of a LOCATE- statement , the
programmer can construct a record within the buffer, referenc-
ing the record using the variable and pointer specified in
the LOCATE- statement. The record is written when either
another LOCATE-statement or a WRITE-statement is executed
for the file. Closing the file after executing the LOCATE-
statement also causes the record to be written.
The REWRITE-statement can only be used with update files;
it causes a record in the dataset to be replaced. The FROM-
option can be omitted if the preceding operation on the file
was a READ with the SET-clause; in this case the record just
read (which is assumed to have been modified) replaces its old
copy in the dataset. Otherwise the FROM-clause specifies the
source of the replacement record, and the KEY-clause, if given,
specifies which record is to be replaced. If the KEY-clause
is not given, then the file must be sequential, and the record
at the current position in the file is replaced. For the
replacement to be acceptable, the current position must be
well-defined, and the preceding operation on the file must not
have been a DELETE.
133
The DELETE- statement, like the REWRITE-statement , can only
be used with update files. If a key is specified, then the
record with that key is deleted. Otherwise the behavior is
similar to that of REWRITE: the record at the current position
is deleted; the current position must be well-defined; and the
preceding operation must not have been another DELETE .
134
STREAM INPUT-OUTPUT
Stream input-output differs from record input-output in
that a transformation is performed when information is moved
between a generation of storage and a dataset. On output, the
transformation consists of translating a data value (which
must be of a printable type) into a character representation
of that value; on input, the transformation goes in the opposite
direction. The character representation need not represent
the value directly, since formatting conventions can be used.
For instance, the character sequence "4 387" could represent
the value 4 3.87 on either input or output, were an appropriate
format to be used. The GET-statement is used for stream
input, while the PUT-statement is used for stream output.
Each of these statements has three variants: list-directed,
data-directed, and edit-directed . With the exception of a few
pathological cases, input is the inverse of output. Thus, if
information is written by a PUT-statement and later read by a
GET-statement of the same form, the original values in storage
will be unchanged. However, it is not in general true that
reading from a file and then writing what was read will yield
the contents of the original file.
The file to be operated on by a GET-statement or a PUT-
statement can be specified either explicitly or implicitly.
The statement
PUT FILE (TABLES) LIST (A, B) ;
causes the printable representation of the values of A and B
135
to be written onto the file TABLES. If the FILE-clause is
omitted, the standard input file SYSIN is assumed for GET-
statements, and the standard output file SYSPRINT (a print
file) is assumed for PUT-statements . >
The stream input-output statements can be used to encode
and decode strings in storage by means of the STRING-clause.
For example, the statement
GET STRING (STR) LIST(A,B);
causes the values of the variables A and B to be "read" from
the string STR, just as though STR was a sequence of characters
on an input file. Similarly,
PUT STRING(STR) LIST(A,B);
causes the values of A and B to be converted to their printable
representations, and then causes the sequence of representations
(with an intervening blank) to be assigned to the character
variable STR.
Line and page skips can be specified in the GET-statement
and the PUT- statement , although certain forms are applicable
only to a PUT-statement that designates a print file. For
instance,
PUT SKIP (2) LIST(A,B);
causes two lines to be skipped before A and B are printed, while
PUT PAGE LIST (A, B) ;
causes a new page to be started before A and B are printed.
The statement
136
PUTLINE(2) LIST{A,B);
causes the printer to be positioned to the second line on the
page before A and B are printed; it differs from SKIP in that
it selects an absolute page position rather than a page posi-
tion relative to the previous line. PAGE and LINE can only
be specified for print files. SKIP on an input file causes
the remainder of the input line to be skipped.
Data Lists
A GET-statement or a PUT-statement can contain one or
more data lists, specifying the items to be read or written.
In its simplest form, the data list is merely a sequence of
scalars, e.g.
PUT LIST (RATE, TIME, RATE*TIME) ;
As this example shows, expressions as well as variables can
appear in the data list of a PUT-statement. The it6ms in
the data list of a GET-statement, however, have the same
restrictions as the targets of an assignment-statement, since
one cannot read a value into an expression. A data list can
also contain iterations, e.g.,
PUT LIST((I,F(I) DO I = 1 TO FNLIMIT) ) ;
Since any item can itself be an iteration, iterations can be
nested to any depth. The DO that controls the statement is
subject to the same restrictions as the specif ied-DO-statement.
A GET-statement can use an input value as an iteration
count, e.g.,
137
GET LIST( (N, (COST(K) DO K = 1 TO N) ) ;
Moreover, aggregates can be included in the data list, e.g.,
DECLARE MIX(40,40) FIXED;
PUT LIST( (MIX(*,I) DO I = 1 TO 40));
In this example the elements of MIX will be printed out with
the first subscript varying most rapidly, as in Fortran. On
the other hand, the statement
PUT LIST (MIX) ;
with MIX declared as above will print out the elements of MIX
with the last subscript varying most rapidly (since this is the
implicit order in storage, as defined by the rules of PL/I).
List-Directed Input-Output
The GET LIST-statement reads an unformatted sequence of
items from the input stream. The data to be read consists of
a list of items separated by blanks or commas, e.g.,
23,47, 'DAVID DAVIS'
Since the dataset is left positioned after the last character
that was read, it is possible to read items from the same line
using several separate GET LIST-statements in succession. The
successive items on the dataset are assigned to the successive
items in the data list. If the data list contains an aggregate,
then enough items are read from the dataset to fill the aggre-
gate. The dataset can also indicate empty items by means of
two commas in a row. For instance, the statement
138
GET LIST(M1,M2,M3) ;
applied to the input
22, ,891
will cause Ml to be set to 22, M3 to be set to 891, and M2 to
be left unchanged.
The types of the items read from the dataset need not
agree with the types of the items in the data list; if there
is any disagreement, the item read is converted to the type
of the item in the data list. For instance, the input value
for a float variable can be written as an integer. A
character-string item on the dataset need not be quoted
unless it contains a comma or blank or starts with a quote.
Thus
THIS ISN'T BAD
can be read into a list of three character variables, and the
variables will receive the strings exactly as written.
However, if the dataset contains the item
'ISN' 'T'
the usual rules for interpreting a character-string constant
will be applied and the receiving variables will receive the
value
ISN'T
The PUT LIST-statement writes a sequence of items onto
the specified output file. Successive items arc -d at
successive tabstop positions, and when a line is filled (as
139
defined by the linesize for the dataset) a new line is
started. If the output file is not a print file, however,
items are separated by single blanks rather than placed at
tabstops. Thus the effect of
PUT SKIP LIST((N DO I = 1 TO 10));
might be to print the lines
12 3 4 5
6 7 8 9 10
under some appropriate assumptions about the linesize and
tabstop positions. Since a PUT LIST-statement does not force
an end of line, several PUT LIST-statements can place output
onto the same line.
Data-Directed Input-Output
The GET DATA- statement reads a sequence of variable names
and associated values, e.g.,
A=3 B=12 D=0;
The pairs in this sequence can be separated by either blanks
or commas (as with the GET LIST-statement) , and the sequence is
ended by a semicolon. The GET DATA-statement itself can, but
need not, specify a list of variables, e.g.,
GET DATA(A,B,C,D,E) ;
or
GET DATA;
The second form is easier to use, but it has the disadvantage
140
that it forces a complete symbol table to be included in the
compiled program. The items in the input stream need not be
given in the same order as the variables in the list. More-
over, variables in the data list can be repeated or omitted
in the input stream. Subscripted and name-qualified variables
can also be included in the input stream, although name
qualifications must be complete. For instance,
A (3, 7) = 'JOE', A (4, 7) = 'SAM', ST.COUV = 19.3;
is a valid input line, assuming an appropriate data list.
The PUT DATA-statement writes a list of variables,
together with their values, onto the output stream. As with
the GET DATA-statement, the PUT DATA-statement need not
contain a list. Execution of the statement
PUT DATA;
causes the values of all printable variables to be written
onto the output stream. Otherwise, the listed variables
are written out. If any of these variables are aggregates,
the scalar elements of the aggregate are written out,
using fully-qualified names and appropriate subscripts.
Unlike the other two forms of the PUT-statement , the
PUT DATA-statement cannot include expressions in the data
list, since expressions do not have names. If the output
file is a print file, successive items are placed at succes-
sive tabstops. Thus, assuming appropriate stored values and
tabstops, the effect of the statements
141
DECLARE
1 AGG(3)
2 (RED, BLUE) FIXED DECIMAL(2);
PUT DATA(AGG) ;
is to print out the lines
AGG.RED(l) = 4 AGG.BLUE(l) = 8 AGG.RED(2) = 0
AGG.BLUE(2)= -5 AGG.RED(3) = 7 AGG.BLUE(3) = 1;
Edit-Directed Input-Output
Edit-directed input-output is accomplished through the
GET and PUT EDIT-statements . For edit-directed input-output,
the transformation between the internal and external forms
of the data is governed by a format list. When an item in
a data list is read, a format is obtained from the format
list and used to transform the character representation of
the item, as it appears in the input stream, into a stored
value. When an item in a data list is written, a format is
used to transform the value of the item into a sequence of
characters to be inserted into the output stream. For example,
the statement
GET EDIT (I, J) (F(4),F(3));
applied to the input stream
^]625.00]6)6]63.1A
causes the variables I and J to be assigned the values 15
(obtained from the first four characters of the stream) and
3 (obtained from the next three characters of the stream) .
Similarly,
142
PUT EDIT(25, 3.142) (2F(7,2));
causes the characters
]6]625.00]6]6\63.14
to be placed into the output stream. In this case, the 2 in
front of the format item F(7,2) indicates that the item is
to be repeated twice.
The available formats are listed in Table 12. There
are two kinds of formats: control formats and data formats.
Control formats control positioning of the dataset; they
cause skipping of information on input, and generation of
blanks, linemarks, pagemarks, and carriage-returns on output.
When a format list is interpreted, control formats are executed
as they are encountered; a control format does not use up an
item from the data list. A data format, on the other hand,
requires a corresponding item in the data list, and the
format item together with the data item determines what action
is to be taken.
When the data list is exhausted, interpretation of the
format list ceases. For instance, execution of the statement
PUT EDIT (J) (E(9),SKIP);
does not cause the control format SKIP to be executed, since
after J is paired with E(9) the data list is exhausted and
execution of the statement is complete. If the format list
is exhausted while items still remain in th' ' ' i list, then
interpretation of the format list starts over again from the
143
Table 12. Format Types
(1) Data Formats
A(u) alphanumeric with field width u
(w can be omitted on output)
B(w) bitstring with field length w
Bl(u) (w can be omitted on output;
B2(u) 31 indicates base 2, 32 indicates base 4,
B3(w) B3 indicates base 8, and
B4 (u) B4 indicates base 16)
F(w,d,s) fixed with field width w, d digits to right
of decimal point, scaling s
{d and s are optional)
E{w,d,s) float with field width u, d digits to right
of decimal point in mantissa, s digits in
mantissa (d and s are optional)
P pia pictured according to picture pia
C(fl,fl) complex with real part formatted using fl,
imaginary part formatted using f2
{fl assumed the same as fl if f2 omttted)
(2) Control Formats
X(w) blank or ignore field of width w
COLUMN (n) continue reading or writing at column position n
TAB(n) skip n tabstops
SKIP(n) skip n lines
LINE(n) position at nth line of printed page
PAGE start new printed page
(3) Remote Format
R(ref) use format obtained by evaluating
reference r-ef
144
Table 13.
Examples
of
Input
Formats
Input Field
Format
Value
TWOfe$CATS
A(8)
TWO)!JCAT£
1011
B(4)
'lOll'B
ONE
B(3)
none —
CONVERSION
raised
)z{)z$1011)z5
B(7)
'lOll'B
2437
33(4)
•OlOlOOOlllll'B
2437
F(4)
2437
lz{2437JzJ
F(6)
2437
)z52437)z$
F(6, 1)
243.7
.2437
F(4,l)
.2437
2437
F(4,l,2)
24370
CATS
F(4,l,2)
.
none —
CONVERSION
raised
2437E1
F(6)
none --
CONVERSION
raised
2437
E(4)
2437E0
2437E-2
E(7)
24.37E0
2437
E(4,l)
243. 7E0
2437E-2
E(4,l)
2.437E0
2.437E-2
E(5,l)
243. 7E0
2437E-2
E(4,l,l)
2.437E0
WrfM
F(3)
0
Wb$
E(3)
0
W2M>1
P ' (6)Z'
2437
2431W
P ' (6)Z'
none --
CONVERSION
raised
2437
C(F(2))
24+371
-2437
C(F(3,1)
F(2) )
-2.4+371
145
Table 14.
Examples of
Outp'
at Formats
Value
Format
Output Field
DYNASTY
A
DYNASTY
2.48
A
}z{)zJ2.48
DYNASTY
A(6)
DYNAST
DYNASTY
A(10)
DYNASTYl!5)z5JzJ
•lOllOOl'B
B
1011001
'lOllOOl'B
B3
544
■IIOIOOI'B
B4
D2
•IIOIOOI'B
B4(5)
D2Jz$)z$)zi
'IIOIOOI'B
B4(l)
D (and STRINGSIZE raised)
2.48
F(6)
]^\6\6)^)62
2.48
F(6,l)
W]ii2.5
2.48
F{6,3,1)
)z50.248
-25
F(2)
none — SIZE raised
24.86
E(13)
]6]6]62.48eE+001
24.86
E(13,0)
]6]6)6\ii2A8eE-002
-24.86
E(13,l,3)
)z5M)z5-2 4.9E+0 00
2.48
P •$$$V.99
1
\6$2.48
17.9+61
C(F(6,1))
]6]611 .9W\66.0
-17.9+61
C(F(6,1) ,F(6))
]6-n .9]6WW^
146
beginning. For instance,
DECLARE A{*) FLOAT;
PUT EDIT (A) (SKIP, 6E(14,5));
causes the contents of the array A to be written out with
six values on a line. As many lines as necessary are used.
A format list can contain repeated items and repeated
lists. The repetition counts and parameters in a format list
are not limited to constants; they can be arbitrary expres-
sions, as in the format list
( (M)E(14,11-K) , (N) (A(L) ,X(5) ,A{L+1) ))
Repetition counts that are not integer constants must
be written in parentheses.
An edit-directed statement can have more than one pair
of data lists and format lists, e.g.
PUT EDIT(A(*,3)) (SKIP,10E(14,DECPT) )
(A(*,4)) (SKIP,6E(14,DECPT) ;
The behavior of input formats applied to various data values
is illustrated in Table 13. Except for the P-format and the
C-format, all of the input formats specify a field width,
which is the number of characters to be read from the input
stream. (The P-format and C-format specify the field width
implicitly.) For the B-formats, the E-format, and the F-format,
the numerical value to be read can have leading and trailing
blanks, which are ignored. Linemarks within a field are
ignored, so that a field can be split over two lines. For the
F-format and the E-format, the second parameter, if present,
147
indicates where an implicit decimal point should be inserted
if none appears explicitly in the input. For the F-format,
the third parameter specifies scaling to be applied to the
input value, while the third parameter of an E-format is
ignored.
The behavior of output formats is shown in Table 14.
The output formats have behavior inverse to the input formats .
The rules are not entirely inverse, however, since the input
formats assign meaning to certain fields that cannot be pro-
duced by the corresponding output formats. For instance, an
output F-format cannot produce trailing blanks, but an input
F-format can read trailing blanks. Moreover, unsealed output
fields always have explicit decimal points unless they are
integers, but unsealed input fields may have implicit decimal
points. Another difference is that on output, the A- format
and the B-formats need not contain field widths, since a field
width can be deduced by converting the output value to character
type or to bit type as appropriate.
A remote format can appear in a format list, either as
the only item in the list or in combination with other items.
The format item R{ref) is interpreted by evaluating the reference
ref, which must be format-valued, i.e., it must evaluate to the
statement-name of a FORMAT-statement . The contents of that
FORMAT- statement are then interpreted. For instance, the effect
of the statements
GET EDIT(A,B,C) (F ( 3) , R (FF) , F { 2) ) ;
FF: FORMAT (F( 5) ) ;
148
is the same as that of the statement
GET EDIT(A,B,C) (F ( 3) , F ( 5) , F ( 2) ) ;
Remote formats are useful when the same format is to be used
in a number of different PUT EDIT or GED EDIT-statements.
The same format can be used for both input and output.
149
BIBLIOGRAPHY
1. American National Standards Institute, "American
National Standard - Programming Language PL/I,"
Report ANSI X3. 53-1976, New York.
2. Beech, D. , A Structural View of PL/I, ACM Computing
Surveys (2)1, March 1970, pp. 33-64.
3. Beech, D. , and Marcotty, M. , Unfurling the PL/I Standard,
SIGPLAN Notices (8)10, October 1973, pp. 12-43.
4. Frieburghouse, R. , The MULTICS PL/I Compiler, 1969 Fall
Joint Computer Conference (35), AFIPS Press, Montvale,
New Jersey, pp. 187-199.
5. Honeywell Information Systems, Inc., "PL/I Language
Manual," Cambridge, Massachusetts, 1974.
6. IBM Corporation, "PL/I Language Specifications,"
Number GY33-600 3-2 , 1970.
7. Lucas, P., and K. Walk, On the Formal Definition of PL/I,
Annual Review in Automatic Programming (6)3, 1969,
Pergamon Press, pp. 105-181.
8. Pollack, S., and Sterling, T. , "A Guide to PL/I,"
Second Edition, Holt, Reinhart and Winston, New York, 1976
9. Radin, G., and Rogoway, H., NPL: Highlights of a New
Programming Language, Communications of the ACM (8)1,
January 1965, pp. 9-17.
150
This report was prepared as an account of
Government sponsored work. Neither the
United States, nor the Administration,
nor any person acting on behalf of the
Administration:
A. Makes any warranty or representation,
express or implied, with respect to the
accuracy, completeness, or usefulness of
the information contained in this report,
or that the use of any information,
apparatus, method, or process disclosed
in this report may not infringe privately
owned rights; or
B. Assumes any liabilities with respect to
the use of, or for damages resulting from
the use of any information, apparatus,
method, or process disclosed in this
report .
As used in the above, "person acting on behalf
of the Administration" Includes any employee
or contractor of the Administration, or
employee of such contractor, to the extent
that such employee or contractor of the
Administration, or employee of such contractor
prepares, disseminates, or provides access to,
any information pursuant to his employment or
contract with the Administration, or his
employment with such contractor.
151
Live Music
Archive
Librivox
Free Audio
Metropolitan Museum
Cleveland
Museum of Art
Internet
Arcade
Console Living Room
Open Library
American
Libraries
TV News
Understanding
9/11